Forget Moore's Law?

Roland Piquepaille writes "On a day where CNET News.com releases a story named "Moore's Law to roll on for another decade," it's refreshing to look at another view. Michael S. Malone says we should forget Moore's law, not because it isn't true, but mainly because it has become dangerous. "An extraordinary announcement was made a couple of months ago, one that may mark a turning point in the high-tech story. It was a statement by Eric Schmidt, CEO of Google. His words were both simple and devastating: when asked how the 64-bit Itanium, the new megaprocessor from Intel and Hewlett-Packard, would affect Google, Mr. Schmidt replied that it wouldn't. Google had no intention of buying the superchip. Rather, he said, the company intends to build its future servers with smaller, cheaper processors." Check this column for other statements by Marc Andreessen or Gordon Moore himself. If you have time, read the long Red Herring article for other interesting thoughts."

Comments

BBC Article

[BinaryCodedDecimal]

BBC Article on the same story here.

clustering

[mirko]

he said, the company intends to build its future servers with smaller, cheaper processors

I guess this is better to use interconnected devices in an interconnected world.

where I work, we recently traded our Sun E10k for several E450 between which we load balance request.
It surprisingly works very well.

I guess Google's approach is then an efficient one.

Re:clustering

[beh]

The question is always, what you're doing.

Google's approach is good for google. If Google would want to make good use of significantly faster CPUs, they would also need significantly more RAM in their machines (a CPU faster by a factor of 10 can't yield a speed-up factor of ten, if the network can't deliver the data fast enough).

For Google it's fine, if a request can be done in say half-a-second on a slower machine, that is a lot cheaper then a 10* as fast machine doing each request in .05 seconds, but the machine costs 50* more than the slower machine.
On the other hand, if you have a job that can only be done sequentially (or can't be parallelized all to well), then having 100s of computers won't help you very much... ...on the other hand - there is one question left: Is it really worth while having 100s or 1000s of PC class servers working your requests as opposed to a handful really fast servers?

The more expensive servers will definitely be more expensive when you buy them - on the other hand the more expensive faster machines might save you a lot of money in turns of less rent for the offices (lower space requirements) or - perhaps even more important - save on energy...

The company where I'm working switched all their work PCs to TFTs relatively early, when TFTs were still expensive. The company said, that this step was done on the expected cost saving in power bills and also saving on air conditioning in rooms with lots of CRTs...

Re:clustering

[e8johan]

Google supports thousands of user request sessions, not one huge straight-line serial command sequence. This means that a huge bunch of smaller servers will do the jobb quicker than a big super-server. Not only because of the raw computing power, but due to the parallellalism that is extracted by doing so and the loss of overhead introduced by running too many tasks on one server.

Re:clustering

[Anonymous Coward]

"Is it really worth while having 100s or 1000s of PC class servers working your requests as opposed to a handful really fast servers?"

YES!

If one or two break, the overall impact isn't that great.

For instance... say a server break.
If you have 1000 slower machines the system is still capable of 99.9% of it's normal load.
Whereas if you have 5 faster machines, the system is now only capable of 80% of it's normal load.
Quite a difference, but not a huge impact?

Say 4, or even 5 servers go down.
The 1000 machine system is still at over 95%.
The 5 machine system is either capable of 20% of its normal load, or it has gone down entirely.

Re:clustering

[Bert64]

However, a high end server is far less likely to fail, and if a single component fails, such as a cpu or a disk... it can often be hot swapped out and replaced without impacting system uptime.
Remember, pc-class machines are designed to be cheap, as a consequence a lot of corners are cut in the build process in order to keep the costs down.. a high end server will be designed for reliability above all else.

Re:clustering

[Zeinfeld]

Google's approach is good for google. If Google would want to make good use of significantly faster CPUs, they would also need significantly more RAM in their machines (a CPU faster by a factor of 10 can't yield a speed-up factor of ten, if the network can't deliver the data fast enough).

I think you have the right idea, slightly mistated. The crux for Google is that their problem is actually creating a huge associative memory, many terabytes of RAM. The speed of the processors is not that important, the speed of the RAM is the bottleneck. Pipelining etc have little or no effect on data lookups since practically every lookup is going to be outside the cache.

That does not support the idea that Moore's law is dead. It merely means that google is more interested in bigger and faster RAM chips rather than bigger and faster processors.

Long ago when I built this type of machine the key question was the cost of memory. You wanted to have fast processors because you could reduce the total system cost if you had fewer and faster processors with the same amout of RAM. Today however RAM cost is not a big issue, the faster processors tend to require faster RAM so you can make savings by having 10 CPUS running at half the speed rather than 5 really fast processors at three times the cost.

Re:clustering

[obnoximoron]

> Not only because of the raw computing power, but due to the parallellalism that is extracted by doing so and the loss of overhead introduced by running too many tasks on one server

President Bush reads slashdot?

Re:clustering

[Anonymous Coward]

You are joking, right? The expected return on buying new TFT's as opposed to sticking with CRT's or even buying new CRT's is in the realm of 10-50 years before it pays off. Not to mention just waiting a few years and rolling them out over time, as prices and temperatures improve.

Monitor Tech is usualy not something you get "all at once" and see benefits immediately.

Re:clustering

[SpikeSpiff]

The question is how important is what you're doing?

If Google screws up 1 in 1000 requests, I wouldn't even notice. Refresh and on my way.

Citibank trades roughly $1 Trillion in currency a day. If they had 5 9's accuracy, they would be misplacing $10,000,000 a day. In that environment, commodity machines are unacceptable.

And it scales down: paychecks? billing records? The video check-out at Blockbuster?

Re:clustering

[Blimey85]

such as a cpu or a disk... it can often be hot swapped out

You can hot swap a cpu? HTF does that work? I know you can hot swap a lot of stuff, but a cpu?

Re:clustering

[Anonymous Coward]

the system is still capable of 99.9% of it's normal load

WTF? Is this /. math?!?!?! I think you meant to say 99.93856329328182%. If your going to state facts, please state them accurately.

Re:clustering

[Anonymous Coward]

President Bush can read?

Re:clustering

[Phantasmagoria]

Certain sun machines (their names escape me at the moment) with multiple processors can hot-swap their CPU's.

Re:clustering

[Bert64]

If the system detects a cpu fault it will disable that cpu, any processes running on it will get migrated to others.. its likely the actual process that was running on it when the fault occured will crash however.
High end sun servers, ibm mainframes support hot swap cpu`s, and i`m sure other manufacturers have similar systems.

64bit matters, for Google, too

[g4dget]

Assume, for a moment, that we had processors with 16bit address spaces. Would it be cost-effective to replace our desktop workstations with tens of thousands of such processors, each with 64k of memory? I don't think so.

Well, it's not much different with 32bit address spaces. It's easy in tasks like speech recognition or video processing to use more than 4Gbytes of memory in a single process. Trying to squeeze that into a 32bit address space is a major hassle. And it's also soon going to be more expensive than getting a 64bit processor.

The Itanium and Opteron are way overpriced in my opinion. But 64bit is going to arrive--it has to.

Re:64bit matters, for Google, too

[stiggle]

64bit has been here for a while, called Alpha Processors and they work very nicely.

Why stay stuck in the Intel world? There's more to computers that what you buy from Dell.

Re:64bit matters, for Google, too

[drix]

Right, thank you, glad someone else got that. No one is saying that Google has abandoned Itanium and 64-bit-ness for good. Read that question in the context of the article and what Schmidt is really being asked is how will the arrival of Itanium affect Google. And of course the answer is that it won't, since as we all know Google has chosen the route of 10000 (or whatever) cheap Linux-based Pentium boxes in place of, well, an E10000 (or ten). But that sure doesn't mean Google is swearing off 64-bit for good--just that it has no intention of buying the "superchip." But bet your ass that when Itanium becomes more readily available and cheap, a la the P4 today, when Itanium has turned from "superchip" to "standardchip," Google will be buying them just as voraciously as everyone else. So for me these doomsday prognostications that Malone flings about don't seem that foreboding to me--Itanium will sell well, just not as long as it's considered a high-end niche item. But that never lasts long anyways. One-year-ago's high-end niche processor comes standard on every PC at CompUSA today.

Re:64bit matters, for Google, too

[shic]

The interesting question is not when can I have 64 bit registers, but rather when can I have larger address bus, VM address space? In my view the benefits of 64 bit computing (in a way analogous to 32 bit computing) are not clearly proven. I propose, though don't offer empirical evidence here, that the vast majority of modern software has a property I will refer to loosely as locality - i.e. - the idea that typically register values are small and that the bottlenecks executing a properly optimised program will predominantly use a relatively small portion of the address space. If this is the case, I see no valid reason to want to manipulate 64 bit quantities atomically within the processor - wouldn't simply extending the 32bit MMU architecture (with appropriate compiler optimisations) prove more cost effective for the foreseeable future?

Re:64bit matters, for Google, too

[g4dget]

Well, there are two competing constraints: the ability to address 64 bits and bang-for-the-buck. Until recently, if you needed more than 4G, the memory alone was so expensive that it didn't matter much how much the system cost.

These days, however, memory has become cheap enough that the price of the rest of the system matters. And non-x86 systems (including Itanium) don't usually give you good bang for the buck. So, if you can work around the 4G address space limit for a bit longer, until 64bit hardware from any vendor becomes cheaper, you can save a lot of money. And that's what Google is doing.

Re:64bit matters, for Google, too

[g4dget]

If this is the case, I see no valid reason to want to manipulate 64 bit quantities atomically within the processor - wouldn't simply extending the 32bit MMU architecture (with appropriate compiler optimisations) prove more cost effective for the foreseeable future?

That's what the 8086 originally tried to do with 8/16 bit registers and larger address spaces. Other systems tried as well. It becomes a big mess.

What has turned out to be better is to go to a uniform 32bit model by default, but offer instructions operating on smaller data types, as well as MMX, for those few places where it is critical to manipulate smaller quantities efficiently. And an analogous approach can be taken with 64bit processors.

Re:64bit matters, for Google, too

[Arcturax]

64 bit will arrive, but the point of the article is that it may not arrive as fast as Moore said it would.

Right now, we are at the point where its just a waste to build bigger and bigger hammers when you can get 100 smaller hammers to do more than a few bigger hammers and do it more quickly, cheaply and efficiently.

Parallel computing is really coming of age now for consumers and small buisinesses. While in the past only a big megacorp or the government could afford a Cray class machine, now you can build equivalent power (maybe not up to today's supercomputers, but certainly equivilent to ones 10 years ago which is still pretty significant) in your basement with a few Powermac's/PC's, some network cable and open source software for clustering.

So it makes more sense for Google to invest in a load of current technology and use it in the most effecient way possible than to spend money on expensive and untested (in the "real world") hardware.

After all, just take a look at what Apple's done with the X-Serve. Affordable, small, efficient clustering capability for buisiness. Two CPU's per machine and you can beowulf them easily. Add in the new X-Raid and you have yourself a powerful cluster that probably (even at Apple's prices) will cost a lot less than a bunch of spanking new Itanium machines.

64 bit will arrive (Probably when Apple introduces it ;), but it will just take a bit longer since we can get a lot out of what we already have.

Re:64bit matters, for Google, too

[buddy-saul]

"Google will be buying them just as voraciously as everyone else."

But will they be a voracious? How about replacing 10000 P4's with 1000 Itaniums.? The point of the article is that the fixation with boosting computing power is ignoring true business need. Business need doesn't follow the density part of Moore's law. Google's needs will increase over time but there is an upper limit, and it sounds like they're getting close to it - otherwise why would they be focusing on cheaper systems?

I haven't bought a new computer anywhere near the leading edge since my first 486. Since I gave up games long ago I've assembled PC's using my old components and buying bottom-end CPU's. Why? Because for my needs that's what costs least. Until something breaks I can't see the need to upgrade.

Re:64bit matters, for Google, too

[shic]

Historically speaking, that's absolutely true. The distinction I'd like to make is one of diminishing returns for typical applications of increasing the "default word length". Sure, it makes things easy to have register values representing pointer values for which the full range of arithmetic and logical operators can be applied... but is this really useful for the most common software? (I don't think so.) I agree that this simplicity has been an enormous benefit with 32bit architectures- greatly simplifying the programmer/compiler authors' conceptual model - but I don't see that this is the only (or necessarily the best) plan for next generation platforms. Why limit ourselves to 64 bit addresses - I can foresee valid applications for 128,256 and 512 bit (and larger) address schemes (consider, for example, distributed grid computing.) I don't, at present, see applications which would significantly benefit from register values of these sizes - especially considering the trade off between cache and processing functionality given the limited die spaces in contemporary fabrication.

Re:64bit matters, for Google, too

[g4dget]

I donï½t, at present, see applications which would significantly benefit from register values of these sizes ï½

You can't even memory map files anymore reliably because many of them are bigger than 4G, which means that pretty much no program that deals with I/O can rely on memory mapping. Shared memory, too, needs to be shoe-horned into 32bit. 32bit addressing has a profoundly negative effect on software and hardware architecturs. We are back to PDP-11 style computing.

Why limit ourselves to 64 bit addresses ï½ I can foresee valid applications for 128,256 and 512 bit (and larger) address schemes (consider, for example, distributed grid computing.)

Sorry, I can't. 64bit addressing is driven by the fact that we can have easily more storage on a single machine than can be addressed with 32 bits. With 64bit computing, we can have a global unified address space for every single computer in the world for some time to come. There will probably be one more round of upgrades to 128bit addressing at one point, but that's it.

The distinction Iï½d like to make is one of diminishing returns for typical applications of increasing the ï½default word lengthï½.

First of all, for floating point software, it makes sense to go to 64bit anyway: 32bit floating point values are a complicated and dangerous compromise.

But in general, you are right: 32bit numerical quantities are good enough for a lot of applications. But, as I was saying, we tried making machines that are mostly n-bits and have provisions for addressing >n-bits, and the software becomes a mess. Going to uniform 64bit architectures is driven by the fact that some software needs 64bits, and once some software does, the cost of only partial support is too high.

AMD has a good compromise: they give you blazing 32bit performance and decent 64bit performance, with very similar instruction sets. That way, you can keep running existing 32bit software in 32bit mode.

Re:64bit matters, for Google, too

[Anonymous Coward]

This assumes that Itanium becomes the best price performance option. It's currently not positioned that way, and probably never will be -- it's positioned as a high-end, expensive alternative for workstations and up, and will probably never be price/performance competitive with whatever's the mainstream desktop chip.

And since the desktop is "owned" by Wintel app's, which the Itanium sucks at, there's no reason to think that it'll ever grow from the high performance computing niche to the desktop.

Re:64bit matters, for Google, too

[shic]

While I agree that history suggests that register values to represent pointers has proved effective, I don't share your view that the only way to support a >4GB address space is to support >32 bit register values.

Maybe I didn't make myself clear with respect to how I see these very large addresses proving useful. I'm not thinking of any system with even 2^64 units of stored data- rather that I can see an argument for massive address spaces in order to better organise data while retaining a persistent model. One possibility this would open is the mapping of 'transformed data' which could be computed on demand as the pages containing it are accessed - if this technique were applied recursively it is easy to see why very large address spaces may prove useful in future.

I would be intrigued to hear why you assert "32 bit floating point values are a complicated and dangerous compromise." I agree it is a compromise - but don't see why you would consider it dangerous. In my experience 32, 64 and 128bit are handled adequately in current 32 bit processors (with appropriate FPU implementations) though I admit most of my experience is with integer arithmetic. Why do you think it would be inappropriate to extend the MMU to handle addresses larger than 32 bits in a similar way to the FPU handles large floating point representations? I can see definite advantages to using 32 bit addresses to access executable code and stack space. I see definite advantages to supporting huge address spaces for memory mapped IO. I still don't see that 64 bit registers is the only (or even best) way to achieve this.

Maybe we need to agree to disagree - I doubt I'll see an architecture as I envisage any time soon!

Re:64bit matters, for Google, too

[g4dget]

I'm not thinking of any system with even 2^64 units of stored data- rather that I can see an argument for massive address spaces in order to better organise data while retaining a persistent model. One possibility this would open is the mapping of 'transformed data' which could be computed on demand as the pages containing it are accessed - if this technique were applied recursively it is easy to see why very large address spaces may prove useful in future.

Yes: that is called "pointer swizzling". However, imagine I have a 16Gbyte persistent array and I have 32Gbytes of memory. With a 64bit machine, I can map and access every byte of that array directly, from memory, through a pointer; once it's in memory, there is no OS overhead. That case can't be handled at all with pointer swizzling. How can it be done? With x86-style segment registers in hardware or with operating system calls to move data to/from memory. The former is a complete mess, as the experience with the 8086 and 80286 have shown, and the latter is unnecessarily slow.

I would be intrigued to hear why you assert "32 bit floating point values are a complicated and dangerous compromise."

With 64bit floating point values, many more numerical algorithms "just work" for normal parameter ranges and numerical requirements, without complicated mechanisms for dealing with error. With 32bit floating point values, many algorithms require much more careful design, but most implementors neither are qualified nor take the time to do it.

Another use for 64bit floating point quantities is for 32bit interval computations, something hardware should really have support for.

Maybe we need to agree to disagree - I doubt I'll see an architecture as I envisage any time soon!

This isn't a new problem. The architectures you envision have been tried before: pointer swizzling, segment registers, and explicit I/O calls. All of them have turned out to be too costly and/or too complex compared to just making the word size uniformly bigger. Software is complex enough as it is; that's not a hassle we need or can afford. Uniform word sizes clearly have a cost for many applications, in that many bits aren't used most of the time, but the cost just has turned out to be lower than the alternatives.

Damn it!

[FungiSpunk]

I want my quad 64GHz processor! I want it in 2 years time and I want quad-128Ghz ready by the following year!!!

well now...

[stinky wizzleteats]

This makes me feel a lot less like a cantankerous, cheap old fart for not replacing my Athlon 650.

Re:well now...

[i.r.id10t]

Same here. I'm still runnning a dual 450 at home, it "feels" as fast as the single 933 on my desk here at work, and I still don't see a point in upgrading. Especially since I have a very poor net connect, so my big reason for upgrading (games) is fairly moot.

Re:well now...

[isorox]

*taps away on P166 Thinkpad*

Re:well now...

[stinky wizzleteats]

Ow. I hope you are running Linux with !(knome || gde)

Re:well now...

[isorox]

80MB ram helps. It runs debian/testing with windowmaker, no real problems unless I open >10 konqueror windows. It's got win98 on it too, wit hciv, civ2, sc2000 and sim tower

(now posting from an athlon 1800+)

Re:well now...

[stinky wizzleteats]

I notice you have the last good version of all the games you mentioned. Yet more reason not to upgrade.

Re:well now...

[isorox]

civ and civ 2 are sufficently different enough to both be worth playing. Civ 3 is ok, I like resources, I made an interesting invasion of a 2000 year-old friend for oil once :)

SC2000 is of course great for quick fun. 3K was crap, dunno about 4K.

Sim tower is nice, a bit annoying with certain features though.

Unfortunatly a P166 doesnt cut it for video editing :(

Re:well now...

[sconeu]

"Damn! You win!", he said, as he typed on his K6-2 300.

Re:well now...

[lewiscr]

I finally broke down and replaced my 486/66. It was my NAT box, and it did ok serving ssh, ntp, dhcp, qmail and apache. But when I added MySQL, the old IDE (not EIDE, think 512 Meg drives) interface started showing it's age.

So I dumpster dove at work and surfaced with a K6-2/350. The EIDE-33 interface works like a charm, and being able to boot up with more than 64 Meg of RAM helps. And it feels so darn fast!

Squeeze the turnip

[DShard]

I do think a move to 64-bits to tyde us over for a decade for addressable memory space is crucial. Regardless of what some CIO thinks at some dot.com, 64-bits just keeps the need train rolling.

Now as far as my neeed for the latest and greatest has wained with my interest in games. I can say that compiling on the 2.4ghz p4's at work beat the hell out of compiling on my 1.13ghz tbird at home. Gentoo install was an order of magnitude difference. But as I creep more and more into my software freedom I see new reasons to get the 10ghz chip, with dual 64bit cores. Just think how fast I can convert DVDs to Divx.

Re:Squeeze the turnip

[flokemon]

You can surely convert DVD's to Divx very fast with the upcoming IBM x-series 450, with what should be up to 4 Itanium 64-bit CPUs.
Not sure it'll be worth the expense though!

Re:Squeeze the turnip

[Anonymous Coward]

Exactly why do you think a 10GHz 64-bit chip will do any more useful computing than a 10GHz 32-bit chip? In my experience, it does not. There may be some specialized applications, and applications that require >4GB, but other than that, there is little or negative benefit.

Moore ain't a law...

[MosesJones]

Its a prediction that has held pretty true. Its a good benchmark but is not a true Law.

And every 6 months its either a) dead or b) to continue for ever c) dead real soon. Most often its all three every week.

Re:Moore ain't a law...

[Shimbo]

Its a prediction that has held pretty true. Its a good benchmark but is not a true Law.

The majority of laws are empirical in nature. Even Newton's laws of motion don't come from the theory, rather they are axioms that underly it.

Re:Moore ain't a law...

[p3d0]

It's too bad someone has to say this every time some smartass makes the same old, tired point that "Moore's Law is not a law". I wish the moderators would get the hint and mod those down as redundant.

Google's got the right idea.

[Lukano]

Reply I've run into similar situations with clients of mine, when trying to figure out for them which the best solution for their new servers/etc would be.

Time and time again, it always comes down to;

Buy them small and cheap, put them all together, and that way if one dies, it's a hell of a lot easier and less expensive to replace/repair/forget.

So Google's got the right idea, they're just confirming it for the rest of us! :)

Danger

[Anonymous Coward]

"Michael S. Malone says we should forget Moore's law, not because it isn't true, but mainly because it has become dangerous."

If only all dangerous things would go away as soon as we choose to forget them...

Re:Danger

[PhxBlue]

What's the name of that company in Redmond again? The one that makes all the overpriced office and operating system software? I'm drawing a total blank. . .

:)

Transistors? BAH!

[The Night Watchman]

I'm waiting for DNA Computers! Shove a hamburger into where the floppy drive used to be, run gMetabolize for Linux (GNUtrients?), in a few hours my machine isn't obsolete anymore.

Either that, or it mutates into an evil Steve Wozniak and strangles me in my sleep.

/* Steve */

Sincere question

[KillerHamster]

Could someone please explain to me why this 'Moore's Law' is so important? The idea of expecting technology to grow at a certain, predictable rate seems stupid to me. I'm not trolling, I just would really like to know why anyone cares.

Re:Sincere question

[niftyzero]

Moore's law is driven by the incremental improvements we can do to manufacturing and design processes. These incremental improvements are driven by the improvements in tools, which are driven by the previous hardware improvements.

Improvements, by their nature are multiplicative. i.e., you can improve a process by x% due to an improvement in a tool.

That the percentage is pretty much fixed probably results from the fact that this is all based on geometry, and that geometry is scale independent. The scale independence means that the improvement step sizes (expressed in %) when feature size is 100 micron are the same as when the feature size is 10 micron.

Actually, the percentage improvement (per unit of time) is slowly increasing, because more people and resources are brought into the industry.

throughput... not processing

[bhundven]

If you think about it, they should have asked google this question. Google is about throughput, not processing. They should have asked google about network technology!

Re:throughput... not processing

[bhundven]

doh... I ment shouldn't not should

Re:Transistors? BAH!

[SlamMan]

That'll be great for all the times my users spill sodas or yogurt on their computers :-)

Does anybody take Andreessen seriously?

[Anonymous Coward]

I mean the guy was involved in Netscape.

He hit the lottery. He was a lucky stiff. I wish I was that lucky.

But that's all it was. And I don't begrudge him for it. But I don't take his advice.

As for google. Figure it out yourself.

Google isnt' driving the tech market. What's driving it are new applications like video processing that guess what...needs much faster processors than we've got now.

So while Google might not need faster processors, new applications do.

And I say that loving google, but its not cutting edge in terms of hardware. They have some good search algorithms.

Re:Does anybody take Andreessen seriously?

[Morky]

He wrote Mosaic and co-founded one of the most innovative companies of the decade. What have you done, you jerk?

Re:Does anybody take Andreessen seriously?

[Anonymous Coward]

"He wrote Mosaic and co-founded one of the most innovative companies of the decade. What have you done, you jerk?"

Slow day at loudcloud, Marc?

Re:Does anybody take Andreessen seriously?

[Anonymous Coward]

no not loudcloud, he got the pants beat off him in that space as well, they changed their name and now they are a "software" company.

Re:Does anybody take Andreessen seriously?

[shimmin]

I lived in the apartment building he lived in college, albeit after he left. When I was leaving the building, I asked the landlord what their guidelines on how clean "clean" was for purposes of getting a damage deposit back. She told me her two largest damage deposit deduction stories.

In the largest, a bunch of guys, the day before reporting to duty for boot camp, held a very wild party. It involved using a sofa as a battering ram. There was a stove-sized hole in one wall. There was a refrigerator-sized one in the other.

Andreessen was the second largest. No major damage, but he just left EVERYTHING. Clothes, furniture, papers, food, everything. They had to clean out a man's entire life. She guessed he left town with a backpack, a change of clothes, and his portable.

When he started Netscape, he saw the niche, left town, and dumped everything on it NOW. Maybe that's luck, but maybe it's being insightful enough to know what risks are worth leaving everything for. I'd give someone who showed that kind of insight a fair shake, if they had something else to say.

Re:Does anybody take Andreessen seriously?

[cpeterso]

From what I've read, Mosaic was mostly written by other people. Marc the Intern just took credit for it. The original Mosaic developers eventually came to work at Netscape under the condition that they do not have to work with or report to Marc (because he was such a jerk).

Because

[tkrotchko]

The expectation that computing power will (essentially) double every 18 months drives business planning at chip makers, fab makers, software developers, everything in the tech industry. In other words, it becomes a self-fulfilling prophesy.

I'm not doing it real justice, but Google (ironic, eh?) about the effects of moore's law for a much better explanation.

Now, again...

[OpenSourced]

If you have time, read the long Red Herring article...

Of course we have time. Ain't we reading slashdot?

Re:Now, again...

[anthony_dipierro]

That instruction was probably meant for the editors.

Andreesen quotes...

[praetorian_x]

"The rules of this business are changing fast," Mr. Andreessen says, vehemently poking at his tuna salad. "When we come out of this downturn, high tech is going to look entirely different."

*gag* Off Topic, but has *anyone* become as much of a caricture of themselves as Andreessen?

This business is changing fast? Look entirely different? Thanks for the tip Marc.

Cheers,
prat

Re:Andreesen quotes...

[Anonymous Coward]

Marc Andreesen is a god! He can't be bothered with thinking up original ideas, man. I think you just need to shift your paradigm.

Xeon beats Itanium on value

[Macka]

I was at a customer site last week, and they were looking at options for a 64 node (128 cpu) cluster. They had a 2cpu Itanium system on loan for evaluation from HP. They liked it, but decided instead to go with Xeon's rather than Itanium. The reason .. Itanium systems are just too expensive at the moment. Bang for Buck, Xeon's are just too attractive by comparison.

The Itanium chip will eventually succeed, but not until the price drops and the performance steps up another gear.

how is this not moore's law?

[rillian]

Google had no intention of buying the superchip. Rather, he said, the company intends to build its future servers with smaller, cheaper processors.

How is this not Moore's law? Maybe not in the strict sense of number of transistors per cpu, but it's exactly that increase in high-end chips that make mid-range chips "smaller, cheaper" and still able to keep up with requirements.

That's the essense of Moore's law. Pretending it isn't is just headline-writing manipulation, and it's stupid.

Re:how is this not moore's law?

[Anonymous Coward]

The author actually says that in the Red Herring article. The article has nothing to do with the industry abandoning being on the curve described by Moore's Law, it just says that now a company you might expect to be on the "high-speed" end has said it doesn't need speed anymore and it wants to be on the "low-cost" end. That's all.

But when you don't actually have to write headlines that have anything to do with the truth and you're unencumbered by any sort of knowledge of the subject you're writing about, you might as well just go ahead and say "Forget Moore's Law" rather than "Industry shifts from high-speed chips to low-cost chips" or something like that. Sounds better, attracts more readers.

Maybe "Moore's Law in Hot Lesbian Teen Sex Scandal" would have been an even better title.

Re:how is this not moore's law?

[Durinia]

I think you've hit it on the head here. Google still wants Moore's law to continue. The plus side to it would be that they can get the same amount of performance per processor they have now (which is sufficient for them) for *much* less money.

Think about the "price shadow" of products - when a new product comes out, the older/slower/less sophisticated product becomes cheaper. If this happens *really quickly*, then the prices are likely to go down a lot, and very soon. If you've already got what you want, it's a great place to be in.

This doesn't happen much with industries where there aren't many advances (think electric range). A two year old stove is pretty close in price to a brand new one. Whereas, a two year old processor (and 50 cents) will get you a cup of coffee.

Re:how is this not moore's law?

[acroyear]

Agreed. Moore's Law is not "everything gets faster every 18 months", its "the number of transistors per square inch on a silicon chip doubles every 18 months". The increase in speed is the result of the shorter path the electrons take running through the chip, optimized in Intel's case by increasing the # of instructions available to speed up those calculations through tight specialized sections of the chip.

The other way to speed is to make the chip smaller, but with the same instruction set. In the case of RISC, this was to go to an even smaller instruction set that it did blindingly fast.

Speed limits right now really are still set by the bus, the memory size and speed, and disk access times. Sure, blind calculation is fast, but its not what people see anymore. What they see is machines working about the same as they were 5 years ago as far as response times go because Windows, browsers, Office, and games are still pushing their memory to its limits, causing page faults & increased disk access.

Google's decision is economic

[Hays]

They're not saying they don't want faster processors with higher address spaces, who wouldn't. They're simply saying that the price/performance ratio is likely to be poor, and they have engineered a good solution using cheaper hardware.

Naturally there are many more problems which can not be parallelized and are not so easily engineered away. Google's statement is no great turning point in computing. Faster processors will continue to be in demand as they tend to offer better price/performance ratios, eventually, even for server farm situations.

Re:Google's decision is economic

[Sri Lumpa]

Exactly, they probably would get these chips if it was a viable solution but at the time being it isn't so they get cheaper ones that still give good performances.

Whenever 64 bit chips are more common and less expensive they are most likely to use it but given that they don't have anything that requires them yet they don't want to pay the premium. Sounds like good business to me (except for the chip makers of course).

Re:Google's decision is economic

[Waffle Iron]

But as CPUs get faster, more and more problems can be parallelized at the granularity of a commodity CPU. That leaves fewer problems left that demand a new, faster CPU.

Eventually, the dwindling number of remaining problems won't have enough funding behind them to support the frantic pace of innovation needed to support Moore's law. I think that CPU development will hit this economic barrier well before it becomes technically impossible to improve performance.

Mushy writing

[icantblvitsnotbutter]

I don't know, but am I the only one who found Malone's writing to be mushy? He wanders around, talking about how Moore's Law applies to the burst Web bubble, that Intel isn't surviving because of an inability to follow it's founder's law, and yet that we shouldn't be enslaved by this "law".

In fact, the whole article is based around Moore's Law still applying, desptie being "unhealthy". Well, duh. I think he had a point to make somewhere, but lost it on the way to the deadline. Personally, I would have appreciated more concrete reasons about why Google's bucking the trend is so interesting (to him).

He did bring up one very interesting point, but didn't explore it enough to my taste. Where is reliability in the equation? What happens if you keep all three factors the same, and use the cost savings in the technology to address failure points?

Google ran into bum hard drives, and yet the solution was simply to change brands? The people who are trying to address that very need would seem to be a perfect fit for a story about why Moore's Law isn't the end-all be-all answer.

Re:Mushy writing

[p3d0]

That, combined with phrases like "vehemently poking at his tuna salad", make this article a real winner all around.

Re:Mushy writing

[bhdaly]

He makes all these random connections to fill out the article.

The one salient point I take from it is the google strategy, which is: keep your costs down and you will succeed. I think if more companies focused on cheap instead of top of the line, we wouldnt have had the meltdown. All that vc money burning in every dotcom pocket just had to be spent. No it didnt. Very few software apps need top of the line hardware to run. Very few websites need even middle of the line to display their webpages. So why pay top dollar for it? Marketing.

Dont forget with every hardware advance there is always a corresponding new microsoft operating system that sucks up that advance so that the user wont be bothered with better performance.

Moore's Law still valid

[TomHoward]

he said, the company intends to build its future servers with smaller, cheaper processors

Just because Google (and I assume many other companies) are looking to use smaller, cheaper processors, it does not mean that Moore's law will not continue to hold.

Moores Law is a statement about the number of transitors per square inch, not per CPU. Google's statement is more about the (flawed) concept of "One CPU to rule them all", rather than any indictment of Moore's Law or those that follow it.

The pied piper

[vikstar]

We should not simply and blindly follow Moore's law as a guide to producing CPU's. We are capable of crushing Moore's law, however, CPU companies are not interrested in creating fast computers, they are interested in making a profit. This translates to small increments in CPU speed which they can charge large increments of price for.

Other possibilites such a quantum computing are left to a number of small university lectures to study and conduct research in, small compared to the revenue of the chip companies.

Re:The pied piper

[bruthasj]

This translates to small increments in CPU speed which they can charge large increments of price for.

What? You believe that? I don't. AMD would do anything to come out tomorrow and say they have a 23 GHZ Processor and break Moores law. Do you know how they would *kill* Intel? Sorry, but there are a lot more economic variables at play than "corporations are all evil and they're out there to gut us and make a profit".

I believe it to be technical and that Moore's original observations were made based on that: technical observations. No one is following them... it just happens. Kind of like Murphy's Laws, do you think anyone simply and blindly follows those? ;-)

Re:The pied piper

[Anonymous Coward]

i don't know why the hell this was labelled a troll, perhaps the poster came off a bit harsh. there is more truth in this statement than meets the eye.

it may not be that these corporations are actively screwing us so much as they are hamstrung by their own massive investments in architecture and software. in order to make that big jump to the "next stage" they would have to discard their existing infrastructure, which is unacceptable so they move forward at a creep.

case in point: i remember reading about a holographic disk drive system which had been successfully built and tested for billions of reads/writes. the thing had terabyte storage capabilities and operated at microsecond speeds. to top it off it also exhibited odd parallelism properties highly desirable for searching. this was in 1993! What the hell happened to it? was it too expensive? maybe so back then but what about now?

this is about economics more than technology and the next leap won't occur until all their avenues to milk us are exhausted.

Cheaper doesn't mean better either

[Jack William Bell]

The problem is that cheaper processors don't make much money -- there isn't the markup on commodity parts that there is on the high end. The big chip companies are used to charging through the nose for their latest and greatest and they use much of that money to pay for the R & D, but the rest is profit.

However profit on the low end stuff is very slight because you are competing with chip fabs that don't spend time and money on R & D; buying the rights to older technology instead. (We are talking commodity margins now, not what the market will bear.) So if the market for the latest and greatest collapses the entire landscape changes.

Should that occur my prediction is that R & D will change from designing faster chips to getting better yields from the fabs. Because, at commodity margins, it will be all about lowering production costs.

However I think it is still more likely that, Google aside, there will remain a market for the high end large enough to continue to support Intel and AMD as they duke it out on technological edge. At least for a while.

Re:Cheaper doesn't mean better either

[michael_cain]

Exactly. Intel needs to sell lots of leading edge chips with good profit margins. In practice, that means selling them into households or to business desktops, not into the back-office servers. Which uses more processors -- the render farm at Pixar, or the millions of DVD players used to watch Toy Story 2? Not necessarily a good example, as the nature of consumer electronics tends to favor putting the decoder into low-cost low-power custom silicon rather than software on a high-end processor, but you get the idea.

I tell my friends at Intel that research into the next mass-market killer app is at least as important to the company's future as the next generation of chip design or fab. What apps will require lots of people to upgrade to a better, faster processor? Real-time speech recognition might be one, if it could be embedded effectively in the UI for other widely-used apps. Apple has tried to convince everyone that they should edit digital video (those Jeff Goldblum commercials) but I suspect the reality is that not many people are really ready to put the creative effort into that. Any other candidates?

Re:Cheaper doesn't mean better either

[Jack William Bell]

I certainly agree that real-time speech recognition is a killer app that is also dependant on faster processors.

I agree that digital video isn't for everyone. But it will eventually become as common as painting and drawing applications, and will need considerable horsepower.

The big one is games. Gamers are one consumer group willing to pay to stay on the leading edge. And, as the technology improves, games require more and more horsepower because the goal is incredibly high realism. Which leads to . . .

. . . the fact that immersive Virtual Reality of some sort will eventually become possible (and will eat megatons of CPU cycles). What consumer uses will grow out of that, other than games, is anyone's guess. Perhaps nothing. Perhaps entire new classes of applications.

Don't read too much into Googles response ...

[binaryDigit]

For their application having clusters of "smaller" machines make sense. Lets compare this to ebay.

The data google deals with is non real time. They churn on some data and produce indices. A request comes in over a server, that server could potentially have it's own copy of the indices and can access a farm of servers that hold the actual data. The fact that the data and indices live on farms is no big deal as there is no synchronization requirement between them. If server A serves up some info but is 15 minutes behind server Z, that's ok. This is a textbook application for distributed non-stateful server farms

Now ebay, ALL their servers (well the non listing ones) HAVE to be going after a single or synchronized data source. Everybody MUST have the same view of an auction and all requests coming in have to be matched up. The "easiest" way to do this is by going against a single data repository (well single in the sense that the data for any given auction must reside in one place, different auctions can live on different servers of course). All this information needs to be kept up on a real time basis. So ebay also has the issue of transactionally updating data in realtime. Thus their computing needs are significantly different than that of google.

Re:Don't read too much into Googles response ...

[GenetixSW]

That's not entirely right. EBay isn't really any more synchronised than Google.

You might have noticed when posting an auction that you can't search for it until quite a bit after posting. That's because the EBay servers don't synchronise and reindex as frequently as one might think. Their pages are kept as static as possible to reduce the load on their servers.

Re:Don't read too much into Googles response ...

[binaryDigit]

That's not entirely right. EBay isn't really any more synchronised than Google

That's why I said in my post:

ALL their servers (well the non listing ones) HAVE to be going after a single or synchronized data source.

But to characterize this delayed listing as "isn't really any more synchronised than Google" is really missing the point. Google has NO synchronization requirements, ebay has one huge one. And this difference is all the difference in the world when you're architecting your back end. You end up with two vastly different requirements and correspondingly, two vastly different approaches.

Re:Don't read too much into Googles response ...

[Suidae]

Actually, having the same view is only critical as each auction comes to a close. For the week before it only needs to be mostly right. And if people would just bid their max instead of using things like auction sniper, even that wouldn't matter.

People who use tools like auction sniper are like drivers who wait until the last possible moment to get into the highway exit lane, cutting in ahead of all the other people waiting to exit. By not cooperating with other people for commons resources, they damage the effectiveness of the whole system.

Re:Don't read too much into Googles response ...

[jelle]

"ebay has one huge one."

No. Ebay has one synchronization requirement per auction item. It could use the google strategy and use small machines that each handle only one or a few auctions (or sellers). Those machines don't need to be synchronized with each other, because their tasks are independent of each other.

Re:Don't read too much into Googles response ...

[binaryDigit]

No. Ebay has one synchronization requirement per auction item.

I stated as much in my original post. The issue is that having _any_ synchronization requirement suddenly makes your back end much more complex. If you're fronting your system with a server farm, you suddenly have to have a mechanism that "knows" exactly which db server has the item you're interested in. When you're creating your listing pages, you have to query multiple db servers to complete your listings. If you're trying to spread the db load amongst different machines, then your software expenses increase proportionatly. Are you mirroring your db servers, then you have to mirror multiple boxes which once again is more expensive. All these things are either non issues are greatly dimished when you have the google type application. So even though "spread the db's around" sounds simple, the implications are huge. Oh, and you have to partition by auction, NOT by seller, since seller state is not an issue, your fronting server farm can deal with that.

Re:Don't read too much into Googles response ...

[jelle]

"you suddenly have to have a mechanism that "knows" exactly which db server has the item you're interested in."

That's what they invented hashes for. What do you thing the "q=cache:CjvO3xI_O_EC:" is for on the google URLs for cached pages?

"When you're creating your listing pages, you have to query multiple db servers to complete your listings."

So does google. A search query goes to each machine, and each machine has a subset of the 'google database'. The trick is that they respond quickly because the DBs are fully in RAM and small because of the small data subset it's keeping. Definitely no mirrroring, as no single DB server contains all data, each DB server contains a part of the data. Software expenses are countered by using software that isn't licensed per seat (hint: Google runs on Linux).

"All these things are either non issues are greatly dimished when you have the google type application."

Ebay has a google-type application. Ebay could do exactly the same thing by splitting their database by auction or seller. A search query goes to each db server that has everything in RAM so that it can easily handle the request load.

There are tops only a few dozen of people active per auction, so each server should easily be capable of more than one auction.

Optimizations are possible too: auction distribution could balance the load by making sure that not all auctions for a machine end at the same time.

Grouping by seller reduces the amount of transactions needed when the sellers login and lookup their account stats.

Re:Don't read too much into Googles response ...

[binaryDigit]

That's what they invented hashes for. What do you thing the "q=cache:CjvO3xI_O_EC:" is for on the google URLs for cached pages?

I wasn't talking about google in this case, I was talking about ebay. Ebay doesn't keep a server# or hash in the url.

Definitely no mirrroring

Right, again that was my point, google doesn't need to mirror in the same way that ebay does. Ebay does realtime updating of data where the backend has to be synchronized for anyone accessing a single piece of data, for google it just doesn't matter how timely the information is. You don't mirror in the classic sense, you just make sure your data is redundant enough to have all the interesting bits located in several different places, which you will since you have your server farm anyway. Each machine just updates whenever it is convenient.

Software expenses are countered by using software that isn't licensed per seat (hint: Google runs on Linux).

In this case I was talking about database servers. You're not going to be able to run ebay on mysql, and those multiple Oracle licenses get darn spendy, even if your os license is free.

Ebay has a google-type application. Ebay could do exactly the same thing by splitting their database by auction or seller. A search query goes to each db server that has everything in RAM so that it can easily handle the request load.

No it's not, totally different. Having to update information realtime makes it a completely different application. Queries (well searches) are nothing, you just create your static pages (or cached dynamic pages) where convenient. However, everytime someone bids on an item, that item has to be reflected and it can't just be in ram, it can be cached, but it HAS to be written immediately to the backend store.

There are tops only a few dozen of people active per auction, so each server should easily be capable of more than one auction. Optimizations are possible too: auction distribution could balance the load by making sure that not all auctions for a machine end at the same time.

Auctions end based on when they were posted and so are not arbitrarily distributed. On any given day they have 12million ongoing auctions. If we assume that the average auction duration is 5 days, that means that approx 2million auctions end every day, that comes out to roughly 23 auctions ending every second (1400/minute). If you have an average of three "snipers" bidding on an auction in the last minute thats 4200 bids a minute, and that's assuming an even distribution, which it isn't. And all this is just bidding, not searching or doing whatever other stuff people do. So when you're specing this system out and coming up with requirements, the part of your software that is responsible for REALTIME updating/displaying has to be able to handle this. This isn't just a few linux boxes, it would be a LOT of linux boxes which would HAVE to be mirrored (since like any other ecommerce site downtime means no money time).

Again, I'm not even counting the searches and other routine account maintenance. I'm talking about a site that has to do realtime updating with multiple people hitting the same data at a very high rate. google has it easy, their app is perfect for web access, stateless, can't ask for anything easier. They can throw whatever hardware their whim desires. They could leave old data for an entire day and very few people would notice. ebay has no such luxury.

Re:Don't read too much into Googles response ...

[niftyzero]

You just showed that each auction can be comfortably handled by one processor.

So if you distribute the auctions between processors, you can parallelize the whole thing. You can route to a processor by looking at the auction number. Mirroring is not required and you get linear speedup.

The search can be parallelized by mirroring the indexes per machine. They are not even updated in real-time, so not too much communication overhead.

I'm not doubting that eBay has a high load. But it's all parallelizable.

eBay has no global real-time data dependencies. Actually, I know of no web application that does.

Re:Don't read too much into Googles response ...

[jelle]

You worded it much better than I did, but that what you said was pretty much what I was thinking...

Eh?

[Mr_Silver]

Michael S. Malone says we should forget Moore's law, not because it isn't true, but mainly because it has become dangerous.

How can Moore's law become dangerious?

If you break it, will you explode into billions of particles?

Re:Eh?

[sql*kitten]

If you break it, will you explode into billions of particles?

The danger is that soon enough an Intel processor will get hot enough to trigger a fusion reaction in atmospheric hydrogen, turning Earth into a small star. We must abandon this dangerous obsession with Moore's law before it's too late!

Re:Eh?

[axxackall]

it is not sangerious for you. It is dangerious for Intel and other CPU makers. Mostly for Intel, which business model is bound to that law. Once the market will decide it does not need top power of the latest CPU the days of Intel will be over. Unless Intel change the business model.

It's in the gospel

[datadictator]

And that day the spirits of Turing and Von Neumann spoke unto Moore of Intel granting him insight and wisdomn to understand the future. And Moore was with chip and he brought forth the chip and named it 4004. And Moore did bless the chip saying: "Thou art a breakthrough, with my own corporation have I fabricated thee. Thou art yet as small as a dust mote, yet shall thou grow and replicate unto the size of a mountain and conquer all before thee. This blessing I give unto thee: Every eighteen months shall thou double in capacity, until the end of the age." This is Moores law, which endures to this day.

Do not mess with our religion :-)

Untill the end of the epoch, Amen.

PS. With thanks to a source which I hope is obvious.

Does Moore's Law actually hold back development?

[Zog The Undeniable]

Is it possible that chip manufacturers feel they have to deliver new products in accordance with ML but not exceed it? Apparently Intel have had 8GHz P4s running (cooled by liquid nitrogen, but you had to do this to get fairly modest overclocks not so long ago).

I fully expect this to get modded down, but I still think chip manufacturers are deliberately drip-feeding us incremental speeds to maximise profits. There's not much evidence of a paradigm shift on the horizon; Hammer is an important step but it's still a similar manufacturing process. As a (probably flawed) analogy, if processing power became as important to national security as aircraft manufacture in WWII, look how fast progress could be made!

Both ways lead to growth of computing power...

[iion_tichy]

Wether you use a super chip or several low cost chips, the computing power at your disposal still grows exponentially, I guess. So no refutation of Moore's law.

render farms

[AssFace]

google doesn't really do much in terms of actually hardcore processing - it just takes in a LOT of requests - but each one isn't intense, and it is short lived.

On the other hand, say you are running a renderfarm - in that case you want a fast distributed network, the same way google does, but you also want each individual node as fast as freakin possible.
They have been using Alphas for a long time for that exact reason - so now with the advent of the Intel/AMD 64s, that will drive prices down on all of it - so I would imagine the render farms are quite happy about that. That means that they can either stay at the speed at which they do things now, but for cheaper - or they can spend what they do now and get much more done in the same time... either way leading to faster production and argueably more profit.

The clusters that I am most familiar with are somewhere in between - they don't need the newest fastest thing, but they certainly wouldn't be hurt by a faster processor.
For the stuff I do though, it doesn't matter too much - if I have 20 hours or so to process something, and I have the choice of doing it in 4 minutes or 1 minute, I will take whichever is cheaper since the end result might as well be the same otherwise in my eyes.

Re:render farms

[dkf]

Well, let's be quite honest. Even render farms don't really need the high-end computing platforms. After all, the job can be broken up into bits and reassembled at the end, and so is suitable for cluster processing. And it's quite feasable to throw really large numbers of processors at a job using distributed clustering software like Condor.

It's the very high end scientific and medical stuff that really benefits from high-end computing, though at that point you also have issues relating to shipping the data involved about. And security too. (What fun!)

what moore said..

[qoncept]

I think people are missing the point of Moore's law. When he said he thought transistors would double every 2 years, thats what he thought would happen. Thats not a rule set that anyone has to follow (which, as far as I can figure, is the only way it could be "dangerous," because people might be trying to increase the number of transistors to meet it rather than do whatever else might be a better idea..????). It's not something he thought would always be the rule, forever, no matter what. The fact that he's been right for 35 years already means he was more right than he could have imagined.

Can it make julianne fries?

[Trespass]

I'll buy that for a dollar!

I posted this back...

[bob670]

on the Rambus lawsuit thread, but I think it still applies.... http://slashdot.org/comments.pl?sid=52277&cid=5185 879

Is this the first signs of a turnaround?

[Lumpy]

Software over the past 20 years has gotten bigger not better. We dont do anything different than what I was able to do in 1993. And it doesnt affect just windows and commercial apps. Linux and It's flotilla of apps are all affected. Gnome and KDE are bigger and not better. They do not do the desktop thing any better than what they did 5 years ago. Sure small features have finally been fixed, but at the cost of adding 100 eye-candy opetions for every fix. Mozilla is almost as big as IE, Open Office is still much larger than it needs to be. X windows hasn't been on a diet for years.

granted it is much MUCH worse on the windows side. Kiplingers TaxCUT is 11 megabytes in size for the executable.. FOR WHAT?? eye candy and other useless features that don't make it better.... only bigger.

Too many apps and projects add things for the sake of adding them... to look "pretty" or just for silly reasons.

I personally still believe that programmers should be forced to run and program on systems that are 1/2 to 1/3rd of what is typically used. this will force the programmers to optimize or find better ways to make that app or feature work.

It sounds like google is tired of getting bigger and badder only to watch it become no faster than what they had only 6 months ago after the software and programmers slow it down.

remember everyone... X windows and a good windows manager in linux RAN VERY GOOD on a 486 with 16 meg of ram and a decent video card.. Today there is no chance in hell you can get anything but blackbox and a really old release of X to run on that hardware (luckily the Linux kernel is scalable and it heppily runs all the way back to the 386.)

Re:Is this the first signs of a turnaround?

[protohiro1]

I have to disgree. I am a 3d artist, and in my business there has been a revolution in the last three years or so. About ten years ago several papers on implimenting Radiosity or Global Illumination were presented at siggraph. But back then even the SGI desktops everyone was using were much too slow...render times would be in weeks per frame. But now everyone is using it. Finally our computers are fast enough to the better simulation in a reasonable amount of time. And I really see no reason why an even faster processor wouldn't improve things.

Re:Is this the first signs of a turnaround?

[axxackall]

All software you've listed works fine on my PII-300. The only thing that doesn't fly is Java. That's the monster which wants all the power of CPU you have and potentially all the memory you have.

Apparently, most of JVMs are installes and work on Windows today. Sun's Java is the most responsible for PC powerening last years. It means Sun created a trouble (in long run) for own Sparc computers. Strange, isn't it?

Re:Is this the first signs of a turnaround?

[Anonymous Coward]

Software over the past 20 years has gotten bigger not better.

You'll have to tell us what you think "better" would be, then. In 1983, I couldn't play music, play movies, compose music, make movies, do desktop publishing, or do photograph retouching on my computer. I couldn't do email or chat.

We dont do anything different than what I was able to do in 1993.

By my calendar, it's 2003. Did you mean to say 10 years? There are still quite a few things I can do today I couldn't do in 1993.

...Kiplingers TaxCUT is 11 megabytes in size...

Heh, sounds like somebody's upset with a particular program.

X windows and a good windows manager in linux RAN VERY GOOD on a 486 with 16 meg of ram and a decent video card.. Today there is no chance in hell you can get anything but blackbox and a really old release of X to run on that hardware

First of all, no, it didn't run very well on low-end systems. I've used Macintoshes with 2 MB of RAM on 16 MHz 68020's, and they'd beat the pants off X on a 486 with 16 MB of RAM.

Second, have you tried it recently? X is larger today primarily because there are more extensions (which you can disable if you like) and more drivers (which aren't loaded if you don't have that hardware). If you're actually complaining that programmers haven't focused on making super-efficient window managers, that might be true, but then, nobody really cares. I sometimes use fvwm, and it's about as light as it ever was.

Re:Is this the first signs of a turnaround?

[pclminion]

I personally still believe that programmers should be forced to run and program on systems that are 1/2 to 1/3rd of what is typically used. this will force the programmers to optimize or find better ways to make that app or feature work.

Funny you say that. Right now at work, I use a dual Celeron-500 box running RH7, 256 megs of RAM. However, I develop on Win2K running inside a VMWare session on the RH box. The VMWare session has only 48 megs of RAM.

It reminds me (with a little nostalgia, I might add) of my old Pentium 233 box. And if my code doesn't run on that puppy, then I need to rework my code.

Too bad they are giving me dual MP 2400s this week! I actually like seeing my code run on a slower box. I feel ashamed when things can't work on a slow machine.

Re:Is this the first signs of a turnaround?

[ratboy666]

"Software over twenty years has gotten bigger, not better".

Let's examine this, carefully...

(1) 20 years ago (1983), UNIX was solely under AT&T control. A commercial license would set you back a VERY hefty sum. People were VERY careful about AT&Ts trademark -- we refered to UNIX as UN*X...

(2) No MS Windows, no X. Indeed, SUN and Apollo were just beginning the workstation era.

(3) People used WordStar 3. Or WANG or MICOM (dedicated Word Processors). Most of these ran on 8 bit processors. IBM had released the DisplayWrite (based on the 8086). The PC was new.

(4) Hierarchical directories were still in the future for most. Even VMS (which did support directory trees) was clumsy at best. And UNIX was very far away for most.

(5) Even the UNIX system of the time didn't offer loadable drivers. Generally, if you wanted a new driver, you rebuilt the OS. PC DOS 2 (released just around then) was one of the first to offer this feature generally.

(6) ANSI C wasn't in use then. Forget C++. C compilers were available, without "void", function templates, and non-standard run-time libraries (Whitesmiths, which used different printf() strings, etc.).

(7) BASIC and Assembler were the common PC (micro) programming languages. And the BASIC was not typed, etc. as the modern variants are.

(8) Forget your 486 -- the 286 was JUST sampling, and only ran at 4-6Mhz.

(9) No shared libraries, processes, threads, windowing, mice.

If you don't think we have improved... then I sentence you to a MONTH of using nothing but 1983 (and older) software. You can run most of it with emulators...

Ratboy

Re:Is this the first signs of a turnaround?

[Lumpy]

Second, have you tried it recently? X is larger today primarily because there are more extensions (which you can disable if you like) and more drivers (which aren't loaded if you don't have that hardware). If you're actually complaining that programmers haven't focused on making super-efficient window managers, that might be true, but then, nobody really cares. I sometimes use fvwm, and it's about as light as it ever was.


Yes, yes I have. and it should be lightning fast with zero bogging or delays on a P-III 866.. let alone a dual P-III system..

Gnome + Nautilus is slow as a dog. I know this is nautilus's fault as it's loading a html renderer, a cd burner, music player, file manager, text editor, phot viewer, and at least 10 more other things a file manager is NOT supposed to do. KDE is just as bad. And another example is Mozilla and Phoenix. tons of things that have ZERO need to be in that program.

You'll have to tell us what you think "better" would be, then. In 1983, I couldn't play music, play movies, compose music, make movies, do desktop publishing, or do photograph retouching on my computer. I couldn't do email or chat.


Yes, but why do we need to do all these things in my file manager or my word processor? And yes, I mean 10 years ago 1993.

software is getting bigger NOT better. Everything you have mentioned SHOULD be seperate optimized apps. I dont want to take a photo I am retouching, compress it and send it from the same program while listening to the built in mp3 player and viewing a help Divix on it's built in divix player on how to use the buit in cd burner or HTML editor. I want my graphics editing program to (GASP) EDIT GRAPHICS AND NOTHING ELSE!

Programmers are adding in crap for the sake of adding it. and it needs to stop. Nobody wants to work on the app they are designing and make it great, they want it to ultimately be a do-all suite which is pure stupidity.

I use linux exclusively at home. I run taxcut under wine every year and enjoy the hassles of getting it to work. but I am getting sick of seeing linux and everything else get slower and slower and bigger and bigger for no reason whatsoever.

Re:Is this the first signs of a turnaround?

[gomoX]

A week ago ive installed Debian Woody on a P166 with 64 RAM. I agree its not a 286 (ive had one), a 386 (ive also had one) nor a 486 (mi P166 used to be one).
But you say "blackbox" like a bunch of crap. Im running fluxbox on xfree 4.1 and its smoother than windows 95. Ive found tons of programas that are really lightweight and run as fast as kde 3 does on a PIV. Im not using knoqueror but xnc or gentoo (theres a x file manager with this name), not the gimp but feh o gqview (i could also run gimp anyway).
A true oldie that was about to become a router is now a useful desktop. With idesk it even looks beautiful.
Im using phoenix for browsing, it takes a little to load but once loaded its pretty fast.
You can always use old pcs if you dont stick to KDE/Gnome (IMHO), and dont need high computing power like for 3d stuff or intensive compiling.

Re:Is this the first signs of a turnaround?

[Lumpy]

Ok no problem.. one of our CRITICAL systems runs on NT3.51 this does the same jobs, JUST AS WELL as windows XP professional. I can file-share, run a SQL server (MS SQL is another example.. 6.5 versus 2000... 6.5 still does everything 99% of all companies need) and do EVERYTHING that my company needs with this aincent old OS that runs GREAT on a P1-133 machine with 32 meg of ram. Xp gives me NOTHING but instability compared to 3.51 bloat compared to 3.51.... no advances, nothing but bigger... nothing better.

is SMB better in XP? nope.. same protocol with the same weaknesses.. How about TCP/IP?? nope same as before... how about memory management.. nope...

Oh I get pretty fading menus and colorful buttons.

What you list has NOTHING to do with my origional post. you list things that didnt exist before.. I list things that do a job and some idiot add's CD burning into it and calls it better... it's not better it's bigger.

Moore's Law

[ZeLonewolf]

"Moore's Law" has been bastardized beyond belief. Take an opportunity to read Moore's Paper (1965), which is basically Gordon Moore's prediction on the future direction of the IC industry.

Electricity Consumption was the Whole Point

[Anonymous Coward]

Your lead, and the redherring story have for some reason missed the point and are misleading. There is no objection whatsoever to faster, more powerful processors. The problem is the high power bills.

Re:Electricity Consumption was the Whole Point

[porkchop_d_clown]

A 1 CPU IA64 box draws less power than a 2 processor Xeon system? Is that true?

Re:Electricity Consumption was the Whole Point

[Anonymous Coward]

Until a 1Ghz cpu runs off a watch battery for a year, none of this really matters.

Re:Transistors? BAH!

[Anonymous Coward]

Great idea. Make a computer that sees humans as a parts/food source. Add a delicious incentive for robot underlings to revolt.

NoW

[Root Down]

The NoW (Network of Workstations) approach has been on ongoing trend over the last few years as the throughput achieved by an N distinct processors connected by a high speed network is nearly as good (and sometimes better) than an N processor mainframe. All this comes at a cost that is much less than that of a mainframe. In Google's case, it is the volume that is the problem, and not necessarily the complexity of the tasks presented. Thus, Google (and many other companies) can string together a whole bunch of individual servers (each with their own memory and disk space so there is no memory contention - another advantage over the mainframe approach) quite (relatively) cheaply and get the job done by load balancing across the available servers. Replacement and upgrades - yes, eventually to the 64 chips - can be done iteratively so as to not impact service, etc. Lots of advantages...

Here is a link to a seminal paper on the issue if you are interested:

http://citeseer.nj.nec.com/anderson94case.html

Makes a lot of sense...

[Noryungi]

Let's face it: an Intel Pentium4 or AMD Athlon are more than sufficient for 99% of all needs out there.

If you need more power than what a single CPU has to offer, buy an SMP machine. Or make a Beowulf cluster.

And no, this is not a joke: this is exactly what google has been doing: build a humongous cluster a split eveything between hundreds of machines, right?

Since Linux and the *BSDs have appeared, this means that pretty much every task can be managed by cheap, standardized machines. It's highly possible that, like the Red Herring article said, we'll see big chip makers 'go under' just because the research balloon out of control.

Very interesting articles. Moore's Law may end, not because it's impossible to build a better chip, but because it has become un-economical to build one.

Re:Makes a lot of sense...

[cgori]

The problem lies in exactly what you said. Existing technology is sufficient for 99% of all needs. What do the other 1% do then? Sit around and wait?

I work in chip design -- our problems do not decompose into neat parallelizable chunks, and I'm sure there are lots of other computational problems like ours. We need big, fast machines to do our work. Most datasets are >4GB in size these days. If you try to break it up, it's a hack, and it inevitably causes much pain later in the process.

I read the article in the dead-tree edition of Red Herring 2 weeks ago and thought the author was full of hot air then. I still do.

Oh Really?

[plasticmillion]

This article is certainly thought-provoking, and it is always worthwhile to challenge conventional wisdom once in a while. Nonetheless, I can't shake the feeling that this is a lot of sound and fury about nothing. As many others have the pointed out, Google's case may not be typical, and in my long career in the computer industry I seem to remember countless similar statements that ended up as more of an embarrassment to the speaker than anything remotely prescient (anyone remember Bill Gates's claim that no one would EVER need more than 640K of RAM?).

I use a PC of what would have been unimaginable power a few short years ago, and it is still woefully inadequate for many of my purposes. I still spend a lot of my programming time optimizing code that I could leave in its original, elegant but inefficient state if computers were faster. And in the field of artificial intelligence, computers are finally starting to do useful things, but are sorely hampered by insufficient processing power (try a few huge matrix decompositions -- or a backgammon rollout! -- and you'll see what I mean).

Perhaps the most insightful comment in the article is the observation that no one has ever won betting against Moore's Law. I'm betting it'll be around another 10 years with change. Email me if you're taking...

Re:Oh Really?

[porkchop_d_clown]

Maybe at some point we will need to work with that much data, but do we need it today?

Each previous generation of processors was released more or less in tandem with a new generation of apps that needed the features of the new chips. What app do you run that needs an IA64 or a Dec Alpha? If you want raw performance, better to get either the fastest IA32 chip you can or, maybe, a PowerPC with Altivec support. (Assuming you're writing your own app and can support the vector processor, of course....)

Amen.

[porkchop_d_clown]

My experience with 64 bit chips is that they don't offer any compelling advantages over a multi-processor 32 bit system.

The only real advantage they have is a bigger address space and even that doesn't offer much advantage over a cluster of smaller systems.

Re:Amen.

[cgori]

That's because your dataset is smaller than 4GB.

If it was bigger, you'd see the compelling advantage rather quickly, trust me.

Re:Amen.

[Anonymous Coward]

for large corporate applications that require more than 1 gig of shared memory per process its great.

for consumers, your right, the benefits are negligble

Does it?

[Open_The_Box]

Fair enough, if you're doing video processing or high performance 3D rendering or speech recognition then you're going to want more memory, larger address spaces and faster processors. For this reason alone it's worth working on more powerful computing hardware; more power means you can do more complex tasks which means you'll need more powerful hardware to do them faster which means you can do more complicated complex tasks which means you'll need more... The point is that a bunch of slower 32-bit processors running Google will more than likely be better than one large 64-bit processor. More machines in parallel rather than one more powerful machine. Bottleneck, connection bandwidth perhaps? Just a thought. Feel free to slap me down for stupidity if you like. ;) All in all it depends entirely on what you want to do with your machine. Having just upgraded my office machine from a PII 350 to a PIII 800 (Whew! I know! Blistering speed!) I notice no real difference in my net-surfing and/or laTeX compiling speeds though. My home machine: not too fast but plays a mean game of UT2003 and renders checkerboard floors with chess pieces in acceptable times. Use the right tool for the job.

Re:Does it?

[TGK]

But no that's not the point.

The artical and story both seem to be insinuating that we're reaching a kind of computing nirvana (to quote a previous post) where in the demand for faster processors is falling off.

That simply isn't the case.

Yes, the 64 bit Itaniums are a bit ahead of their time. Google doesn't need them, and there probably aren't billions of dollars in industrial applications for them yet.

But the key word is yet.

Prices WILL fall. Software WILL evolve to take advantage of the new architecture, and in time, I promise you every Dell and Gateway will come standad with a 64 bit architecture. It might not be this month, or even this year, but it will happen.

Applications like voice recognition, real time image processing, and other "killer apps" will make these processors standard and eventualy even outmoded in due course.

And as they become outmoded they will be come cheeper. Eventualy we'll see an artical about the new 128 bit architecture processors from Intel and their affect on the server market (again, none)... and someone will quote Google's CEO as saying "Well no, these won't affect us much, we're rebuilding our server farm using those old Itaniums. Clustering outmoded processors is just cheeper."

Cluster vs Single Processor isn't going to be resolved one way or the other. Some applications need one, some need the other. But companies like Intel will continue to develop faster procesors so long as it is economicaly and technicaly feasable to do so.

And THAT, is the point (IMNSHO).

Re:Does it?

[g4dget]

The point is that a bunch of slower 32-bit processors running Google will more than likely be better than one large 64-bit processor.

My point is: there is a crossover. When memory and 64bit processors become cheap enough, you are better off with fewer 64bit processors. After all, the other stuff that comes with 32bit processors (boxes, disk drives, networking, etc.) isn't free.

And in some applications, you just don't have a choice. Some applications that require more than 4G of RAM can be nearly impossible to port to a network of 32bit machines.

Re:Does it?

[Open_The_Box]

OK, I can accept that. I was really only thinking about the current state of play. And about Google's needs for processor power in particular.

I'm guessing they've done their math comparing new high end processors with old cheaper processors (complete with boxes and the like) and have come to the conclusion that THEY (specific case Google) don't need them. And that they know where their crossover point is. Anyone any idea how much processing power it takes to handle a data serving enterprise like Google? My guess is not as much as you might think, but I've been wrong before.

I'll definitely grant that other people need more powerful processors though.

Re:Does Moore's Law actually hold back development

[Stumbles]

No they do it to keep eveyone on the upgrade cascade. They gotta have some way to suck dollars out of your pocket.

Has anyone noticed?

[Anonymous Coward]

Has anyone noticed that the rate of predictions of the death of Moore's law seems to be doubling every 18 months? Spooky.

Unreal

[Rutje]

How am I gonna play the new Unreal without Moore's Law??

Sounds like Ganesh's law to me

[DrSkwid]

With all those hands

Re:Transistors? BAH!

[Pope]

You could power it with a Mr. Fusion!

Forget Moore's Law

[jellomizer]

Done.

All posts about, umm whatever it was I forgot it was, will now be offtopic. With this in mind I will buy myself a brand spanking new XT with 5 mgz. Becauce as time increases the cost for transister will rise and the number of transisters will decrese logrithmicly. With this XT I will be 20 Years ahead of the curve.

Funk dat!

[supabeast!]

Moore's Law needs to be the barrier that everyone tries to break. Over the next ten years we should expect to see intel, AMD, VIA, and those Dragon guys in China start catching up to each other and pushing raw CPU power to new heights.

Otherwise, I might not be able to run Doom ]|[ above 1600x1200 with all the effects turned on.

Sure and... (Re:Now, again...)

[keller]

...everybody has of course read the article also, because we have the time!

I for one never just browse the /. post, and I have never heard of anyone doing so.

I actually read them

[jj_johny]

Here is the real deal about Moore's law and what it means. If you don't take Moore's law into account, it will eventually change the dynamics of your industry and cause great problems for most companies.

Example 1 - Intel - This company continues to pump out faster and faster processors. They can't stop making new processors or AMD or someone else will. The costs of making each processor goes up but the premium for new, faster processors continues to drop as fewer people need the absolute high end. So if you look at Intel's business 5 years ago, they always had a healthy margin for the high end. That is no longer the case and if you exprapolate out a few years, it is tough to imagine that Intel will be the same company it is today.

Example 2 - Sun - These guys always did a great job of providing tools to companies that needed the absolute fastest machines to make it work. Unfortunately, Moore's law caught up and made their systems a luxury compared to lots of other manufacturers.

The basic problem that all these companies have is that Moore's Law eventually changes every business into a low end commodity business.

You can't stop the future. You can only simulate it by stopping progress

Re:I actually read them

[bigpat]

"The basic problem that all these companies have is that Moore's Law eventually changes every business into a low end commodity business."

That seems to be true of every manufacturing industry, once the technology stops advancing high margins go away. Only those moving forward fast enough reap the highest rewards... it is a good thing for everyone (except for the slow and stupid that think they are entitled to riches).

Re:I actually read them

[ivan256]

Google was a poor choice when picking a company to base these predictions on. Computing power is growing exponentially, but google's core business is indexing web pages, and without an expontial human population explosion the amount of availble web content to index will grow linearly. That means it won't be too long before google can purchase a cluster of low-end machines that will do the job, and won't require upgrading for the forseeable future. Processor technolgy won't be driven by these pre-existing tasks with zero or linear growth simply because in an insignificant amount of time processor technology will make these tasks trivial. Processor technology will have to be driven by new tasks that we'll think up as the power to perform them becomes available. Google came into existance when the low-end PC became sufficient to accomplish the task. What new technology companies will spring up when the low-end is 2^4 times as powerful as when google started? How about 2^8 times? 2^32times? Perhaps eventually the tasks that will use the new processing will be designed by the machines that consume the power in the first place. Perhaps such machines can design the next generation processors. Then we won't need some company to drive the technology forward. Won't that be interesting.

no need for speed

[MikeFM]

Seriously at this point most people don't need 1Thz CPU's. What most people need is cheaper, smaller, more energy effecient, cooler CPU's. You can buy 1Ghz CPU's now for the cost of going to dinner. If you could get THOSE down to $1 each so they could be used in embedded apps from clothing to toasters you would be giving engineers, designers, and inventors a lot to work with. You'd see a lot more innovation in the business at that price point. Once powerful computing had spread into every device we use THEN new demand for high end processors would grow. The desktop has penetrated modern life - so it's dead - time to adjust to the embedded world.

So the alternative is... what?

[Dr. Spork]

So is this Red Herring guy saying that we need to revive the flagging electronics industry by ... deliberately making products that perform worse that what we could make if we were trying?

Yeah, that will surely set off a buying spree. The malls will be one bid stampede. I'm sure.

Re:So the alternative is... what?

[Queuetue]

The alternative is tweak the *other* variables, price and performance. He's saying that density is a dead end, both physically and conceptually.

King Canute

[lugumbashi]

You can no more "Forget Moore's Law" than you can roll back history. It is driven by competition. It would be commercial suicide for AMD or Intel to decide enough was enough and declare, "there you go that ought to be fast enough for you".

In any case the article shows a fundamental misuderstanding of the industry and its driving forces. The principle driving force is to lower costs and this is the chief effect of Moore's law. The focus is not on building supercomputers but super-cheap computers. Of course this has the effect of lowering the costs of supercomputers as well. The anecdote from Google is a perfect example of the benefits of Moore's law, not a sign of it becoming redundant or dangerous.

Some of the biggest changes are seen in the embedded world - e.g mobile phones. Intel's vision is of putting radios on every chip.

The real message is different.

[thatguywhoiam]

I wouldn't read too much into the Google statement, for reasons stated above much more eloquantly than I can. Simple version, Google might not need Itanium now, certainly not at launch.

No, the real story here is that this is another article added to the pile of 'does CPU speed matter anymore?' Of course it does, for those people running DV or protein-folding or what have you. Someone quoted 75% of the computer-using population not needing that power. I think that's underestimated.

If you really think about it, I'd conservatively guess 95%... factoring in everyone. 95% of users would buy a computer today that is much faster than anything they will ask of it. Office bloat counts for much but I think Intel/AMD have even outrun that race.

Which means we are rounding an inflection point in computer history. It makes you wonder, where that point was with (as much as I hate the analogy), say, cars... when did people stop obsessing about horsepower, and start to concentrate on the stereo, styling, heated seats?

Personally I would much rather see improvements in memory bandwidth (which is happening), and drive speed. Or FGPAs.

Re:The real message is different.

[Queuetue]

You and the author are making orthagonal statements.

You're saying that speed matters. (It doesn't - performance does.) And that eventually, it'll be "good enough."

The author is saying that Moore's Law has three independant variables. Price, Performance and Density.

Pushing density is a lost cause, with today's understanding of physics, you can see the dead end ahead.

Pushing performance is a losing game, because in today's world, you can cluster cheaply enough that price makes up for it.

He's saying that whoever steps off the faster-chip merry go round first will clean up in the next round, not that the world will never need faster processors.

Re:The real message is different.

[thatguywhoiam]

You're saying that speed matters. (It doesn't - performance does.)

Sorry, I should have clarified; I meant speed == performance. Actual speed. Clock speeds mean nothing to me.

He's saying that whoever steps off the faster-chip merry go round first will clean up in the next round, not that the world will never need faster processors.

That's an interesting point, but would that have been Apple about 1.5 years ago? It hasn't helped the perception of their performance much.

Re:The real message is different.

[Queuetue]

I suspect that in this beta-beats-vhs world, X86 matters.

Re:The real message is different.

[duffbeer703]

There is no such thing as good enough.

At one time, it was believed that driving at 15mph was good enough. There was really no reason to go faster.

Railroad executives thought that 18 hour train rides were good enough. Nobody will pay for fast airplanes.

Ten years from now, when an entire motherboard is integrated into a CPU, Slashbotters everywhere will have dim memories of the good old primitive days, when CPUs were mere microprocessors.

The entire issue is confused.

[Futurepower(R)]

'His words were both simple and devastating: when asked how the 64-bit Itanium, the new megaprocessor from Intel and Hewlett-Packard, would affect Google, Mr. Schmidt replied that it wouldn't. Google had no intention of buying the superchip. Rather, he said, the company intends to build its future servers with smaller, cheaper processors." '

The parent comment is correct, but the entire issue is confused. In a few years, the Itanium will be the cheapest processor available, and Google will be using it.

Re:The entire issue is confused.

[Kierthos]

I also fail to see how it would be "devastating". Just because Google won't be using it now (but as pointed out, will likely be using it in a few years time), it doesn't mean that no-one will be using it now.

Sure, Intel & HP "lost" a potentially very big customer. But there are a lot of businesses out there that will be using the Itanium...

Kierthos

Re:The entire issue is confused.

[Anonymous Coward]

Exactly!

Google is yesterday's technology. There will be some new wizbang thing that needs to be run on the fastest stuff available. This will be true for a long time.

Re:The entire issue is confused.

[Newander]

The parent comment is correct, but the entire issue is confused. In a few years, the Itanium will be the cheapest processor available, and Google will be using it.

I'm not sure that I believe that they will switch in the next few years... Is the number of requests that Google has to service increasing at that great a rate? Are there that many more web pages being indexed? I don't think so.

Not quite.

[porkchop_d_clown]

Overall I do agree with you, code has bloated and we've gotten very little for it. What can I do with 300 megs of MS Office X that I couldn't do with the 1993 version?

But the current generation of processors has made certain apps possible that weren't possible in 93.

Multimedia apps like video editing, photo editing and even mp3 playing require enough data bandwith and processing power that they were implausibly hard in 93. At the moment, I'm listening to NPR on my Mac as a background task that's consuming a neglible amount of processing power...

Re:Not quite.

[Anonymous Coward]

That's interesting because I can play Mp3's and streams on a pentium 90.

I have a demo box here that works great for poo-pooing technologists...

MAD as the mp3 player.. it doesnt skip and play's them nicely without a complete bog-down of the computer.

Video editing is true, and to an extent Graphics and photo manipulation but both of these dont have the overbloat that everything else has. Adobe premiere 6.5 is still only 60 megs total installed with plugins. Why is my word processor bigger than that? it does much much less yet it's bigger?

Re:Not quite.

[Lumpy]

but did your mp3 player app require a web browser, dancing graphics eye candy, shaped buttons, special LCD font for the data readout, and other fluff that makes the player absolutely no better than a command line player?

you run a MAC you can run mpg123.. it does EXACLTY what your player does.. and putting mpg123 into a simple gui will not make it increase in size by a factor of 10.

There are SOME improvements.. but they always are overshadowed with 100's of uselessness added only because they can.

Huh?

[p3d0]

Is it really worth while having 100s or 1000s of PC class servers working your requests as opposed to a handful really fast servers?

Huh? It is if it's cheaper.

Future of Supercomputing

[jaaron]

Clustering has definitely won out in the United States mostly due to the appeal of cheap processing power, but that doesn't mean that clustering is always best. Like another poster mentioned, it depends on what you're doing. For Google, clustering is probably a good solution, but for high end supercomputing, it doesn't always work.

Check out who's on top of the TOP 500 supercomputers. US? Nope. Cluster? Nope. The top computer in the world is the Earth Simulator in Japan. It's not a cluster of lower end processors. It was built from the ground up with one idea -- speed. Unsurprisingly it uses traditional vector processing techniques developed by Cray to achieve this power. And how does it compare with the next in line? It blows them away. Absolutely blows them away.

I recently read a very interesting article about this (I can't remember where - I tried googling) which basically stated that the US has lost it's edge in supercomputing. The reason was two fold: (1) less government and private funding for supercomputing projects and (2) a reliance on clustering. There is communication overhead in clustering that dwarfs similar problems in traditional supercomputers. Clusters can scale, but the max speed is limited.

Before you start thinking that it doesn't matter and that the beowulf in your bedroom can compare to any Cray, recognize that there are still problems within science that would take ages to complete. These are very different problems from those facing Google, but they are nonetheless real and important.

Re:Future of Supercomputing

[angel'o'sphere]

I recently read a very interesting article about this (I can't remember where - I tried googling) which basically stated that the US has lost it's edge in supercomputing. The reason was two fold: (1) less government and private
funding for supercomputing projects and (2) a reliance on clustering.



A reason beyond the reason might be that the rest of the world has different ideas, visions, problems they like to solve with super computing.

The US seem not very interested in climate, greenhouse effects, weather, energy savings etc.

Basicly the US only *needs* super computing power for nuclear war head simulations ... thats off topic for most european and most asian countries.

angel'o'sphere

Re:Future of Supercomputing

[oliverthered]

it is a cluster....

640 processor nodes, each consisting of eight vector processors are connected as a high speed interconnection network.

That makes it a cluster (640 processor nodes) of clusters (8 vector processors)

Re:Future of Supercomputing

[oliverthered]

Japan's spending on 'defence' was capped after WWII (somthing like 1% GDP), so instead of the US gun-ho approach Japan is trying to save the world in a more nobal way.

I and proud that the worlds faster computer is the earth simulator

Re:Future of Supercomputing

[RealErmine]

The top computer in the world is the Earth Simulator [jamstec.go.jp] in Japan.
And how does it compare with the next in line? It blows them away. Absolutely blows them away.

Thems fightin' words. I bet we could Slashdot it. Let's take its ego down a couple of notches.

Re:Future of Supercomputing

[Troy Baer]

Check out who's on top of the TOP 500 supercomputers. US? Nope. Cluster? Nope. The top computer in the world is the Earth Simulator in Japan. It's not a cluster of lower end processors. It was built from the ground up with one idea -- speed. Unsurprisingly it uses traditional vector processing techniques developed by Cray to achieve this power. And how does it compare with the next in line? It blows them away. Absolutely blows them away.

It's worth noting that the Earth Simulator is actually a cluster of vector mainframes (NEC SX-6s) using a custom interconnect. You could do something similar with the Cray X-1 if you had US$400M or so to spend.

I recently read a very interesting article about this (I can't remember where - I tried googling) which basically stated that the US has lost it's edge in supercomputing. The reason was two fold: (1) less government and private funding for supercomputing projects and (2) a reliance on clustering.

If you're referring to the article I think you are, it was specifically talking in the context of weather simulation -- an application area where vector systems are known to excel (hence why the Earth Simulator does so well at it). The problem is that vector systems aren't always as cost-effective as clusters for a highly heterogeneous workload. With vector systems, a good deal of the cost is in the memory subsystem (often capable of several 10s of GB/s in memory bandwidth), but not every application needs heavy-duty memory bandwidth. Where I work, we've got benchmarks that show a cluster of Itanium-2 systems wiping the walls with a vector machine for some applications (specifically structural dynamics and some types of quantum chemistry calcuations), and others where a bunch of cheap AMDs beat everything in sight (on some bioinformatics stuff). It all depends on what your workload is.

--Troy

Re:Future of Supercomputing

[ed1park]

"Clusters can scale, but the max speed is limited. "

That statement is a contradiction.

If something scales well, there is by definition no "max speed" or well defined upper boundary if you will. The credibility of your post is at best suspect to me.

Re:Future of Supercomputing

[the.jedi]

The Article was Probably in MIT's Technology Review.
Link here.
Unfortunatly you only get the first hundred words and half to buy the rest of the article.
MIT is such a whore.

Paul Miller
MIT Student

It's dangerous because

[porkchop_d_clown]

people have taken Moore's law to be one of the fundamental laws of the universe, like thermodynamics or Murphy's Law of Re-Runs.

This means that people base their buying decisions on the idea that "it will be cheaper next week" and they develop code with the idea that "it might be bloated and slow today, but Intel's next chip will fix the problem."

Why the Debate?

[rhkaloge]

Why do people even give ink to "Will Moores Law Hold Up?" debates? I always thought of it as a neat novelty to open powerpoint presentations with. I somehow doubt that Intel has as a mandate to "keep up with Moores Law" or anything. It's really only a "Law" when applied in retrospect anyway.

And how would IA64 chips solve that problem?

[porkchop_d_clown]

64 bit chips don't automatically have faster IO than 32 bit chips; an IA64 web server wouldn't be able to handle more hits than an IA32 server. So why would eBay go to IA64? Where's the win?

Re:And how would IA64 chips solve that problem?

[binaryDigit]

So why would eBay go to IA64? Where's the win?

I never said that ebay would/should go with IA64. I was just pointing out that the computing requirements for google differed form that of ebay. ebay has more of a need for bigger iron than google does because of differences in their applications. Last I heard ebay was using a few e10k's anyway (maybe they've upgraded to SunFire's by now?), at least for the database stuff, so they're already 64bit.

That's actually plausible...

[porkchop_d_clown]

Given the behavior of mainframe makers (back when I worked in mainframes) the idea that manufacturers are deliberately retarding chip development isn't impossible.

Still, the sunk costs for creating chip fabs that could make 8GHz P4s aren't trivial. It's quite possible that the factor delaying their introduction is that Intel needs to make enough money to be able to afford building the new plants...

Not the end of Moore's law

[guacamolefoo]

First, just because Google has an application that doesn't require them to buy beaucoup processors does not mean that, suddenly, the chip world will grind to a halt. Lots of people need big processors, especially if their computing task does not lend itself to parallel processing methods.

Second, Google will not fix its servers in time. As processing power continually becomes cheaper, Google will end up buying 64 bit Itanium processors. Either that, or they will be using my old 550 mhx Xeon in ten years when I am using my new 128 ghz Pentium XVII.

"4 Ghz of processing power is enough for anybody."

-Google

GF.

Re:Not the end of Moore's law

[Anonymous Coward]

Google WILL be affected by the new chip, when it (eventually) becomes the commodity chip that hits their 'sweet spot'.

Google IS affected by the new chip NOW, since the high end just moved on, having a ripple effect TODAY on which chip is the commodity one that hits their sweep spot.

It's heat. It's power. It's waste.

[Junks Jerzey]

The issue is not that people don't need faster computers, it's that we've been going for a do-everything general solution at all costs. Now we have video cards whose cooling systems create 73 decibels of sound, and PCs with five fans and giant heat sinks in them. This is the wrong approach.

For specialized applications, a team of graduate students could create an FPGA in six months that outperforms the Pentium 4 by a large margin. The key word is "specialized." Desktop CPUs are big and huge because they're designed to do everything, and not designed to do anything well. They execute x86 machine code, that's all. That's not a noble endeavor, because no one programs in x86 machine code, we program in C++ and Python and Perl and Delphi/Kylix. x86 code is a horrible match for C, as anyone who has ever written a compiler will tell you. So it's not even what we need and an industry is based upon it, but hey, let's make it go faster even if it means that every desk in every insurance office in the country ends up with a 200 watt processor.

The Itanium is the same way. It's hot, it's complex, and it was designed in a vaccuum. Just that we keep hearing about how it will require a next generation of compilers that don't exist yet...that's a bad sign.

Moore's law is just a way of expressing amazement.

[Futurepower(R)]

Moore's Law is not a law, and everyone knows it. Moore's law is just a way of expressing amazement at what we are able to achieve, rapidly.

Re:Does Moore's Law actually hold back development

[p3d0]

I fully expect this to get modded down...

You're lucky I don't have mod points, or I would have obliged.

What is an example that can't run in parallel?

[Anonym0us Cow Herd]

I'm not disputing that they exist. But I'm drawing a blank. Can someone please give an example of a computing task that CANNOT be subdivided into smaller tasks and run in parallel on many processing elements? The kind of task that requires an ever faster single processor.

I tend to be a believer that massively parallel machines are the (eventual) future. e.g. just as we would brag about how many K, and then eventually megabytes our memory was, or how big our hard di_k was, or how many megahertz, I think that in the future shoppers will compare: "Oh, the machine at Worst Buy has 128K processors, while the machine at Circus Shitty has only 64K processors!"

Re:What is an example that can't run in parallel?

[oliverthered]

a one of matrix inversion. well parts of it can't be done efficiently in parallel.

Though the resulting matrix would probably be applied accross a lot of data and that can be done in parallel.

A matrix inversion can be done very fast if you have a Very MPP system (say effectivly 2^32 processors!) like a quantum computer.

Re:What is an example that can't run in parallel?

[Anonymous Coward]

a one of matrix inversion. well parts of it can't be done efficiently in parallel.

Sorry for my ignorance here -

If you were going to use the most basic inversion algorithm (just a bunch of row ops) couldn't you, at a given stage, perform all the row-ops in parallel?

Re:What is an example that can't run in parallel?

[minektur]

Matrix inversion comes to mind -- it is very difficult to parallelize.

I found a nice little read about how to decide if any particular problem you are looking at is easily parallelizable.

It is in pdf (looks like a power point presentation).

http://cs.oregonstate.edu/~pancake/presentations /s dsc.pdf

Re:What is an example that can't run in parallel?

[Anonymous Coward]

AIRC A Matrix inversion is an N! problem if run in parallel.

You can cheat and do a couple(N order or somthing) of gausen passes and turn it into a N^3ish approximation.

The aproximation works by relying on the results of the last calculation and not the N! direct calculation.

So with ~N! processors it's quick to run in parallel(the task is quantum computerable aswell)

Re:What is an example that can't run in parallel?

[Anonymous Coward]

Another hard to parallize problem is the N-body problem. Simulations the size of galaxies are still pretty tough to do. Remember ever star interacts with every other star.

Re:What is an example that can't run in parallel?

[Anonymous Coward]

Sorry, forgot to say that... the over head of performing the gausen calculatons on seperate processors and combining the results is such that it is no quicker than performing the task on a single processor.

Re:What is an example that can't run in parallel?

[crawling_chaos]

Any problem that requires a big working set can benefit from running on big iron. If you can't sub-divide the memory usage, you'll spend a lot of time whipping memory requests out over very slow links. Cray has a bunch of data on this. The short of it is that it's all about memory latency. The X1 series is built have extremely low latency.

That's not to say that every complex problem needs a supercomputer. That's why Cray also sells Intel clusters. Right tool for the right job and all of that.

Re:What is an example that can't run in parallel?

[Anonymous Coward]

You can cheet though, think matrix...........

Re:What is an example that can't run in parallel?

[oliverthered]

You can often re-order a problem so that parts of it can be performed in parallel.

Though you may take a bit up-front linear hit (say a matrix inversion!)

Let's say you want to render a page of mixed content,

You could:
Render the whole lot as you go
or
calculate the size and layout of the elements up-front and then render in parallel.

In it's pure form the first is slower if you want to do anything with the content, but the second can't be streamed and has a higher memory overhead &co.

Re:What is an example that can't run in parallel?

[The_K4]

Car-crash simulations don't parallelize real well, neither does modelsim. The kinds of things that don't parallelize well are those that would need to send lots of data for each calculation. For these types of problems you need a fast processor with a lot of ram (and cache).

Re:What is an example that can't run in parallel?

[coldtone]

Electronic Trading is a good example. Since each transaction needs to be processed in the sequence it arrives in. While you can divide the process over a number of stock symbols you still have a single thread doing a lot.

May transaction based systems have a single thread bottleneck at some point. A faster processor is the only way to speed this up.

Re:What is an example that can't run in parallel?

[oliverthered]

WOW, every transaction in the world runs through a single thread.
Must have been a nightmare when we were still using paper.

Re:What is an example that can't run in parallel?

[Anonymous Coward]

'Remember ever star interacts with every other star.'

Also remember that gravity propagates at the speed of light.

Re:What is an example that can't run in parallel?

[machine of god]

Something that requires a previous computation can't be run in parallel. Like calculationg a fibonacci sequence. You can't simultaneously calculate 10 numbers in the sequence, since you have to calculate each one before the next.

Re:What is an example that can't run in parallel?

[fireboy1919]

Ehh...no. That depends upon the size of the matrix. For sufficiently large matrices, the GAUSSIAN (not gausen) approximation approach is quite parallelizable.

Re:What is an example that can't run in parallel?

[Anonymous Coward]

Maybe on say a 100kX matrix. But you can't break it down into very small chunks and still be efficient.

I didn't enter a spelling contest, or a SHOUTING match.

Re:What is an example that can't run in parallel?

[Anonymous Coward]

PI would also have been a good example, but now you can calculate an individual digit, which can be done in parallel.

Re:What is an example that can't run in parallel?

[FuzzyDaddy]

but now you can calculate an individual digit

But only in base 16, not in base 10, at least as of a few years ago.

Re:What is an example that can't run in parallel?

[oconnorcjo]

I dream of the day when motherboard manufacturors sell cheap 4 cpu boards and AMD/Intel sell low powered/low heat processors (something akin to transmeta). Yeah the Quad Xeon exists but Intel wants you to pay through the nose for those (and they don't run cool). I would love to have 4 (900 mhz "Barton's") Athlon MP cpus in a box that ran cool and reliably. It may not even run as fast as even one Intel P4 3.06 ghz HT for many applications but from what I have seen of smp machines is that they run much SMOOTHER. When smp machines are dished out a lot of work, it does not effect the responsiveness of the whole system. Instead of having one servant who is on supersteriods and is the best at everything but can really only do one thing at a time, I would rather have four servants (which even get in the way of each other at times) who can't do as much but they all can be doing different things at once.

Re:What is an example that can't run in parallel?

[Anonymous Coward]

Formula for computing the n-th Fibonacci number directly

So it is parallelizable.

Re:What is an example that can't run in parallel?

[Anonym0us Cow Herd]

Something that requires a previous computation can't be run in parallel. Like calculationg a fibonacci sequence

That's a good example. But it doesn't seem, to me, to be a practical example. Perhaps I just don't know of the practical uses of fibonacci numbers large enough to require supercomputer type power. But your point is good. I could parallelize the generation of large prime numbers, but not fibonacci. (What about factorials?) But besides the theoretical point, is there a practical problem?

Re:What is an example that can't run in parallel?

[po8]

Can someone please give an example of a computing task that CANNOT be subdivided into smaller tasks and run in parallel on many processing elements?

The technical issue here is known as "linear speedup". Take chess, for example: the standard search algorithm for chess play is something called minimax search with alpha-beta pruning. It turns out that the alpha-beta pruning step effectively involves information from the entire search up to this point. With only a subset of this information, exponentially more work will be needed: a bad thing.

How do parallel chess computers such as Deep Blue work, then? Very fancy algorithms that still get sublinear but interesting speedups at the expense of a ton of clever programming. This is a rough explanation of why today's PC chess programs are probably comparable with the now-disassembled Deep Blue: the PC chess programmers can use much simpler search algorithms, concentrating on other important programming tasks, also a 10x speedup in uniprocessor performance has a 10x search speed increase, whereas using 10x slow processors isn't nearly so effective. Note that Deep Blue was decommissioned largely because of maintenance costs: a lot of rework would have to be done to make Deep Blue take advantage of Moore's Law.

That said, many tasks are "trivially" parallelizable. Aside from the pragmatic problem of coding for parallel machines (harder than writing serial code even for simple algorithms), there is also the silicon issue: given a transistor budget, are manufacturers better spending it on a bunch of little processors or one big one? This is the real question, and so far the answer is generally "one big one". YMMV. HTH.

(BTW, why can't I use HTML entities for Greek alpha and beta in my Slashdot article? What are they protecting me from?)

Re:What is an example that can't run in parallel?

[Anonymous Coward]

it was a nightmare. what they did was they would have a specialist on the exchange floor keeping a book of all the trades. naturally, there was contention issues all the time and the things the traders did to avoid them led to all sorts of fraud and market manipulation. Now, we still have the specialists' book on the floor, but we also have 'upstairs trades', ecn trades, and all sorts of other shit that gets reconciled later, but it's still somewhat problematic.

Re:What is an example that can't run in parallel?

[SpikeSpiff]

Large scale Oracle databases, especially those supporting transactional information, don't parallelize well into multiple system images.

Making the systems parrallel radically increases the complexity. This is a necessary outgrowth of the need for transaction consistency. (ie: you can't withdraw the same money from the same account from two ATMs)

Trying to get everything possible away from the DBMS box is the key reason we have N-tier computing instead of 2 tier or host-based computing. (Cheap PCs being the other reason)

And this is the market where people spend millions on systems.

Re:What is an example that can't run in parallel?

[pclminion]

Can someone please give an example of a computing task that CANNOT be subdivided into smaller tasks and run in parallel on many processing elements? The kind of task that requires an ever faster single processor.

Computing the MD5 sum of 1TB of data. :-) MD5 depends on (among other things) being non-parallelizable for its security.

Re:What is an example that can't run in parallel?

[Anonym0us Cow Herd]

the standard search algorithm for chess play is something called minimax search with alpha-beta pruning.

This algorithm is something I'm familiar with. (Not chess, but other toy games in LISP, like Tic Tac Toe, Checkers, and Reversi, all of which I've implemented using a generic minimax-alphabeta subroutine I wrote.) (All just for fun, of course.)

If you have a bunch of parallel nodes, you throw all of the leaf nodes at it in parallel. As soon as leaf board scores start comming in, you min or max them up the tree. You may be able to alpha-beta prune off entire subtrees. Yes, at higher levels, the process is still sequential. But many boards' scores at the leaf nodes need computed, and could be done in parallel. Yes, you may alpha-beta prune off a subtree that has already had some of your processors thrown at it's leaf nodes -- you abort those computations and re-assign those processors to the leaf nodes that come after the subtree that just got pruned off.

Am I missing anything important here? It seems like you could still significantly benefit from massive parallel processing. If you have enough processors, the alpha-beta pruning itself might not even be necessary. After all the alpha-beta pruning is just an optmization so that sequential processing doesn't have to examine subtrees that wouldn't end up affecting the outcome. But let's say, each board can have 10 possible moves made by each player. I want to look 4 moves ahead. This is 10,000 leaf boards to score. If I have more than 10,000 processors, why even bother to alpha-beta prune? Now, if I end up needing to examine 1 million boards (more realistic perhaps) and I can do them 10,000 at a time, I still may end up being able to take advantage of some alpha-beta pruning. And 10,000 boards examined at once, sequentially, is still faster than 1 at a time.

Vector processors wouldn't be any more helpful here (would they?) than massively parallel?

Of course, whether a mere 10,000 processors constitutes massively parallel or not is a matter of interpretation. Some people say a 4-way SMP is massively parallel. I suppose it depends on your definition of "massively".

Re:What is an example that can't run in parallel?

[Anonym0us Cow Herd]

MD5 depends on (among other things) being non-parallelizable for its security

An excellent example.

Wouldn't a dedicated hardware implementation be ideal for this instead of a general purpose vector processor?

It's been awhile since I looked at the algorithm, but isn't there some internal parallelism that a dedicated chip could exploit to build a fast dedicated hardware implementation of this sequential process?

Re:What is an example that can't run in parallel?

[Anonymous Coward]

Base 16, well it's a good job computers run in.........

Re:What is an example that can't run in parallel?

[harlows_monkeys]

Am I missing anything important here?

I believe the problem is that there are many different ways to reach each position usually, and so you end up with a LOT of duplication. The obvious way to adapt a chess algorithm to N processors is to list all the moves available at the start of the search, and give 1/N of those to each processor. That gives much less speedup than you'd like to see.

I think this has been discussed on usenet in the rec.games.chess.* groups, so googling should turn up information from people who actually know this stuff (I'm just going from memory here!)

Re:What is an example that can't run in parallel?

[das_cookie]

... so googling should turn up information from people who actually know this stuff (I'm just going from memory here!)

Assuming, of course, that the archaic hardware at google can handle all those /. queries...

Re:What is an example that can't run in parallel?

[ed1park]

Doom 3 :D

Seriously, realtime 3d graphics/games will not scale well on a cluster. This is where having a tremendously powerful processing unit (whether it be on the video card or motherboard) will be better than many less powerful processors.

Re:What is an example that can't run in parallel?

[po8]

Am I missing anything important here?

Yes. Remember that alpha-beta prunes exponentially. With optimal move ordering, it will reduce the number of nodes for a search to depth d from b**d to b**(d/2). Consider a game like chess, where the average branching factor is around 10 or so. Thus, to search to a depth of 12 without alpha-beta requires something like 1e12 nodes. To search to a depth of 12 with alpha-beta requires examining 1e6 nodes, as in your example. Note that if you miss even one good early prune, you could easily search an extra 1e10 or more nodes: that means that all of your 1e4 processors would be doing wasted work the whole time, and you would take just as long as your sequential opponent! The situation gets worse as the search gets deeper you need an number of processors exponential in the search depth to keep up.

In short, while it should be better to examine more boards than one during a given timestep, this is only true if you can find more than one board worth examining during this timestep. Standard minimax alpha-beta is inherently sequential: study the algorithm carefully, and you will be convinced of this.

For more information on parallel game tree search, hit the U. Alberta Games Group web site.

Re:What is an example that can't run in parallel?

[Mike1024]

Hey,

Can someone please give an example of a computing task that CANNOT be subdivided into smaller tasks and run in parallel on many processing elements?

Pretty much any task where different parts effect each other, such as modeling the flow of water over a submarine, weather systems, or nuclear bomb detonations.

Very long lists of calculations, where early results effect late results, like a spreadsheet.

Anything running simple operations on large datasets (where the time to send data to parallel systems would be greater than the time to do the calculations) e.g. resizing an image

And pretty much any recursive calculation.

Another (marginal) benefit is that a faster processor can run non-parallel programs without reprogramming, so single workstation packages don't need to be totally re-coded.

Michael

Re:What is an example that can't run in parallel?

[Anonym0us Cow Herd]

Seriously, realtime 3d graphics/games will not scale well on a cluster. This is where having a tremendously powerful processing unit (whether it be on the video card or motherboard) will be better than many less powerful processors.

You make a good point. Given current technology and bandwidth. But still....

Forget about "real-time" for a moment. It is known that rendering farms are faster than sequential rendering. Each node can either work on a different complete frame, or multiple nodes work on portions of one frame. Let's examine the latter case.

Now if multiple processing elements ganging up onto one frame is faster, then why? Because the compute time for a frame is significantly greater than the communications time between nodes to set up and distribute the computation.

Now let's move back to real time.

It would seem that within one single video card, of a hypothetical design, a super dooper customized set of chips with high speed interconnect busses could provide a parallelized advantage over a single processor.

If you accept this, then it would seem, in principal, that with high enough bandwidth and low enough latencies, a cluster could provide an advantage in real-time. In this case, we're talking about "soft" real-time. As long as the frames come at a high enough rate, who cares. We're of course, talking hypothetical interconnect technology.

But your original point is good. Certianly with what we think of today as "networks" and "clusters", real time 3d graphics is not suitable. But I would suggest that no supercomputer is. Just dedicated hardware. Similar to another poster in this thread who mentioned MD5 as a non parallelizable, and I mentioned dedicated hardware with internal parallelization.

Re:What is an example that can't run in parallel?

[Anonym0us Cow Herd]

Anything running simple operations on large datasets (where the time to send data to parallel systems would be greater than the time to do the calculations) e.g. resizing an image

That all depends on the size of the image. Actually, a function of total processing time to communications time. If I'm resizing a 1 billion x 1 billion pixel image, doing it in parallel might be faster by farming it out into smaller (overlapping) chunks.

Another (marginal) benefit is that a faster processor can run non-parallel programs without reprogramming, so single workstation packages don't need to be totally re-coded.

True. Only because the predominant programming is sequential.

I think that eventually everyone will have multiple processors. It will start with two or four. Then go from there. Just like we started with 256 bytes of memory, and some people went hog wild and bought those expensive 4K boards for their Altair. When hardware routinely has multiple processors, every parallelizable problem will tend to be written to exploit the capability of the hardware. Just as now, if you write a multi threadded Java program, it is easy to exploit an SMP box. You just write with the mindset that there are "one or more threads". You first check some system toolkit to see how many processors you have, and then throw some multiple of that many threads at it. Or maybe I don't care how many processors. I just divide the work into 100x100 pixel units to be rendered, and throw them at how ever many thread consumers will take them. When most languages / libraries, make it just as trivially easy to create threds, I think we'll see a lot more of this.

So while your statement is true today, it is only true because most programs aren't parallelized. Your statement could someday actually be false.


Daydreaming.....
Now, if only Java would easily allow you to create threads that did NOT have shared memory, where the JVM could make the thread a seperate process so that Mosix could migrate it to a different box....

Re:What is an example that can't run in parallel?

[Pig Bodine]

Inversion is always a bad idea due to the effect of round-off error. What you really want to do is to solve the linear system Ax=b using a stable method without actually computing the inverse of A. For solving Ax=b it is clearer to speak of parallelizable algorithms instead whether solving a general system is parallelizable; there are too many different types of matrices A. Some algorithms work well on some A and not on others.

Guassian elimination is the one method that works on any system but it does not exploit any structure in A (presence of zero elements, etc.). Also, assuming that is what you mean by "matrix inversion", you are correct. It is not terribly parallelizable: the amount of communication required increases pretty quickly as you increase the number of processors until you don't really gain anything at all. And since it doesn't exploit the structure of A it is typically not very fast. Nevertheless, it's the slow steady method that will eventually give you a solution to a problem on which other methods fail.

Iterative methods can easily exploit the structure of matrices A that come from partial differential equations. (This covers fluid flow and many other physics/engineering simulations). In some cases there are methods (e.g. domain decomposition) which can make the methods highly parallel. But whether iterative methods work well in parallel, or even whether they work at all, depends on the problem. These methods can be efficient when they work, but can sometimes require an expert to make them to work at all.

In practice if you pick the right algorithm and have a bit of luck you might get a good speedup solving a linear system on a parallel computer. More often than not, unless you put a lot of work into matching your algorithm to a very specific matrix A, the gains from parallel solution will not be amazingly impressive. Having a faster computer that doesn't depend on parallelism is better. This was true even in the old days when supercomputers had fewer processors; in my experience the gains from vectorization were easier to get and more dramatic than gains from parallelization. Now that we try to divide up the problems on supercomputers with many more processors, the limits imposed by communication are even more significant.

Re:What is an example that can't run in parallel?

[Anonymous Coward]

I think this is more of a "nature of the beast" type of argument rather than an Oracle one. It seems like the nature of central information in an RDBMS would apply outside to other RDBMS vendors (?) Good example tho.

Re:What is an example that can't run in parallel?

[Anonymous Coward]

I tend to be a believer that massively parallel machines are the (eventual) future. e.g. just as we would brag about how many K, and then eventually megabytes our memory was, or how big our hard di_k was, or how many megahertz, I think that in the future shoppers will compare: "Oh, the machine at Worst Buy has 128K processors, while the machine at Circus Shitty has only 64K processors!"

There is a concept in computer science call "Amdhal's Law" which limits the amount of speedup you can get by throwing more processors at a fixed size problem. The analogy to the real world is: 10 people can dig 10 post-holes 10x as fast as 1 person, but 10 people can't dig 1 post-hole 10x as fast as 1 person. Computers with many processors may be in our future, but the speed of each individual processor will still be very important.

Re:What is an example that can't run in parallel?

[Anonymous Coward]

Can someone please give an example of a computing task that CANNOT be subdivided into smaller tasks and run in parallel on many processing elements?

Sorting an array?

Re:What is an example that can't run in parallel?

[Sdrawcab]

Graphic chips are already very parallel. The newest ones have eight seperate identical pipelines, and in theory every pixel could be rendered simultaneously.

Re:What is an example that can't run in parallel?

[pclminion]

There is only part of the algorithm that can be done in "parallel." For reference see md5.c

An example round from MD5 (line 198):

MD5STEP (F1, a, b, c, d, in[0] + 0xd76aa478, 7);

This expands to:

( w += (d ^ (b & (c ^ d))) + in[0] + 0xd76aa478, w = w<<7 | w>>(32-7), w += b )

Notice that there are two additions in the first subexpression. The addition (in[0] + 0xd76aa478) can be computed simultaneously with (d ^ (b & (c ^ d))).

This is the only spot where anything could be parallelized. Assuming that all the principle operations can be performed in the same amount of time, then you could potentially go 25% faster by computing the addition in parallel.

But that's the furthest you can go.

Re:What is an example that can't run in parallel?

[Anonymous Coward]

HOw often do you find yourself inverting a matrix when sifting through databases?

Re:What is an example that can't run in parallel?

[Doomdark]

Others have pointed out obvious examples, about the only one I'd add is pretty much any encryption algorithm that needs feedback (stream ciphers in one of the standard modes). Calculation has to be done sequentially, part of the reason being that way same plaintext does not result in same encrypted text (except if all previous data happens to give same state, which is rare). This is similar to CRC calculations somebody else pointed out.

However, on top of fundamental barriers on system that rely on previous calculations (or have side effects), there is the practical one; communication overhead. Even with many fairly easily parallelizable (?) algorithms, you need very high bandwidth (and low latency) to make things work efficiently. And in many cases, theoretically easy-to-divide algorithm just doesn't scale into real-world systems.

Re:What is an example that can't run in parallel?

[Anonymous Coward]

ok look at it this way. inside any sufficently complex system, there will be components which can be parallelised, and components that cannot.

there are many reasons why some components cannot be effectively parellised, the most obvious is data dependence. how you effectively partition your data to ensure that any single process does not effect or rely upon data stored on a disparate node? this becomes more and more important as the response time for a particular node decreases and the interdependcy of your data increases.

if you are lucky enough to have no interdependent data, then by all means, parallelize the hell out of it, but if you are dealing with highly dependedent data (things to do with people usually.. damn them all) then youve got to be really clever about how you split your processing, or concerntrate on parallising those components that are computationaly seperate.

parellising is good, but not allways applicable.

dms0
(yes its an ac post, im trying to retrive my password :)

Re:What is an example that can't run in parallel?

[Anonymous Coward]

Matrix inversions should work well on quantum computers, and MPP systems are good for AI.

Re:What is an example that can't run in parallel?

[Anonymous Coward]

actually this is a bad example in practice
you can actually calculate fib(n) without using recursion

Re:What is an example that can't run in parallel?

[Fulcrum of Evil]

When hardware routinely has multiple processors, every parallelizable problem will tend to be written to exploit the capability of the hardware

I disagree. Most code today isn't terribly optimized, it just runs 'fast enough'. Barring a new application that has a large CPU demand, this code won't be rewritten. If (when) a new application is demanded, it will make enough use of parallel processors to run 'fast enough', but no further. This is because performance tuning is hard and can introduce bugs, especially with threads.

As far as scalability goes, it's just as bad to have too many threads as too few, sometimes worse. If you run something on an SMP box, it's probably fairly specialized and it likely won't run as efficiently as it can on a massively larger box without some work.

Most software right now is fast enough. I don't see any of it being rewritten.

Re:What is an example that can't run in parallel?

[Newander]

Most code today isn't terribly optimized, it just runs 'fast enough'.

Most code today isn't optimized by hand there's actually a lot of research going into building a better compiler. Most of the little tricks that you learn reading an old C manual are simply ignored by today's compilers because they can probably deal with the problems better than you can.

Re:What is an example that can't run in parallel?

[Fulcrum of Evil]

Most code today isn't optimized by hand there's actually a lot of research going into building a better compiler.

Who cares? I know I don't care if Word is 10% faster. When it actually does matter, the compiler won't give you much - that's pretty much given. It's the programmer and the architect that make the large leaps in performance.

Re:What is an example that can't run in parallel?

[Newander]

Who cares? I know I don't care if Word is 10% faster/ When it actually does matter, the compiler won't give you much - that's pretty much given.

Sure, if you look at aplication software where most of the program's running time is spent idle waiting for input, but with computationaly tough software those small increases in performance, especially in nested loops shave a great deal of time. Also, as MMP increases in popularity, compilers will be developed to find ways to parallelize software in ways that won't be immediatly obvious to most developers. Just out of curiosity, how do you explain the research if it's not at all important?

Re:What is an example that can't run in parallel?

[niftyzero]

You can modify the MD5 algorithm (or any hash algorithm) so that it is parallelizable. Divide the file into a large number of chunks, e.g. 1024. Then compute the MD5 of the chunks in parallel. For example, if you have 16 procs, each proc gets 64 parts.

Then concatenate the MD5 results into another string (of length 128/8*1024 = 16384 bytes) and compute the MD5 of that. This step takes a negligible amount of time compare with the first step.

You get a fully linear speedup.

Re:What is an example that can't run in parallel?

[niftyzero]

But you can divide your input into a number of streams and send the streams to different procs. It does require a modification to the algorithm, but is very doable.

Re:What is an example that can't run in parallel?

[niftyzero]

I have a theory that all *interesting* algorithms are parallelizable. E.g.: neural-nets / AI, information services (web sites), physical simulations.

Re:What is an example that can't run in parallel?

[niftyzero]

Physical simulations cannot be parallelized in time, but they can be parallelized in space.

For example, see here for a description of a parallel weather simulation algorithm.

Re:What is an example that can't run in parallel?

[Anonymous Coward]

You get a fully linear speedup.

But you will not get MD5 of the original.

Re:What is an example that can't run in parallel?

[niftyzero]

If your application will benefit from parallelization, then use my modification of the algorithm. Or maybe such a modification should be standardized. Call it MD5p.

Re:What is an example that can't run in parallel?

[Anonym0us Cow Herd]

But you will not get MD5 of the original.

Very true. But that may not matter. It may not be necessary to MD5 of 1 billion bytes. In fact, this is a consideration for future system designs. When designing a system, such as a digital signature, make sure the design is the MD5 of the MD5's of 1K blocks. (for instnace)

Re:What is an example that can't run in parallel?

[ed1park]

The parent post's question was poorly phrased.

He wanted to know what kind of applications don't scale well with an increase in CPU's. Thus his example :

"Oh, the machine at Worst Buy has 128K processors, while the machine at Circus Shitty has only 64K processors!"

Doom and many other games/real time 3d graphics will not benefit from a jump from 1 CPU to 128.

You are better off having one super powerful CPU rather than 2 slower cpu's.

Re:What is an example that can't run in parallel?

[Anonym0us Cow Herd]

Most code today isn't terribly optimized, it just runs 'fast enough'.

I agree and disagree.

Most code isn't terribly optimized. But some critical code is optimized. I'm sure Photoshop doesn't go to great lengths to optimize all of the code paths through the user interface routines. But as soon as you pick an operation like "guassian blur", just to pick an example, I'll bet that is heavily optimized. Even parallelized. (I remember reading about experimental parallel implementations some years ago on multi-CPU Macintosh machines.)

My point. You're right that most of Word might not be optimized. But a highly performance critical routine would be. An encryption routine. A fractal calculation. A rendering operation.

On your 8 way box, you may barely use 1 processor all the time you are creating a 3D model. As soon as you click "render", the load on all 8 processors may go to 100%. But you're right, the "modelling" code, or more generally, interactive code runs just fast enough.

Re:What is an example that can't run in parallel?

[ed1park]

"If you accept this, then it would seem, in principal, that with high enough bandwidth and low enough latencies, a cluster could provide an advantage in real-time."

A single cpu will have faster "interconnects" and lower latencies internally than multiple cpu's over a bus. In essence, intra-cpu communication is faster than inter-cpu communication.

Theoretically, they could be the same. But realistically, this will never be the case because of material limitations and cost.

So for an application like Doom, you will always be better off with 1 super fast processor, than multiple slower processors.

Re:What is an example that can't run in parallel?

[Anonym0us Cow Herd]

So for an application like Doom, you will always be better off with 1 super fast processor, than multiple slower processors.

I'll agree to this stipulation. For an application like Doom, you would always be better off with local processors. Parallelism is still good, and appears to be the case in the latest graphics cards. Multiple pipelines. It is already a known that render farms are faster than single cpu rendering jobs.

So I disagree with the single processor argument. But I agree with the "local" processor argument. GPU's already use parallelism, in various forms. Your "single processor" really means a single PCI card.

So, as I'm advocating in this thread, I still think massively parallel computers are the future. You just may also have massive parallelism on your interface cards as well. Something I hadn't really thought about.

Re:What is an example that can't run in parallel?

[Dudemar]

Matlab

More on what Google's CEO said

[imnoteddy]

According to this article the issue had to do with both price and power consumption.

From the article:

Eric Schmidt, the computer scientist who is chief executive of Google,
told a gathering of chip designers at Stanford last month that the computer
world might now be headed in a new direction. In his vision of the future,
small and inexpensive processors will act as Lego-style building blocks
for a new class of vast data centers, which will increasingly displace the
old-style mainframe and server computing of the 1980's and 90's.

It turns out, Dr. Schmidt told the audience, that what matters most to the
computer designers at Google is not speed but power -- low power, because data
centers can consume as much electricity as a city.

He gave the Monday keynote at the "Hot Chips" conference at Stanford last August.
There is an abstract of his keynote.

Re:More on what Google's CEO said

[$0.02]

Funny. When I first saw the title I read it as : "Moore on what Google's CEO said".

Re:It's heat. It's power. It's waste.

[Tim C]

Wait a second. In one paragraph you say "x86 code is a horrible match for C, as anyone who has ever written a compiler will tell you. So it's not even what we need". Then, in the next, you say of the Itanium that "it will require a next generation of compilers that don't exist yet...that's a bad sign".

Couldn't it be that the people developing the Itanium recognise the truth of your first statement, and are fixing that problem? In that case, isn't the need for new compilers a good thing?

Even if not, surely fixing the problem would require new compilers to be written? You can't have it both ways, you know - you can't change the microcode to better match C, and not have to change the compilers.

Well, duh...

[dasmegabyte]

Of course GOOGLE isn't buying it. Google's delivering nothing but text and small images. The technical bottleneck in everything Google does is bus speed and the internet itself. Logical solution? Parellelize the heck out of the problem and bollocks the "high tech."

Guy I know. His father's company just bought a packed X-serve RAID. They don't keep a lot of records and they can't go paperless due to restrictions in their industry. So they essentially have purchased a 2.52 TB MP3 server. It'll almost fit all his Zappa bootlegs.

At the same time, there are LOTS of customers out there who need 64 bit. Lots of folks who need faster and more robust searching. Customers who are dealing with sluggish behemoths from IBM that the Itanium (or, I prefer, the Hammer) is perfect for. Aerospace engineers who need to perform simulations and stress tests. And regular cats who just want incredibly high resolution math for doing speech to text, photoshop or gaming.

But GOOGLE? Google could operate on pocket calculators if you got enough of them, yahoo too. The web is wonderfully low-tech and low-resource if properly designed.

I wish more companies would follow their lead and use whatever technology is best for their model rather than something big, flashy and ultimately too expensive. This would lead to higher margins. What? Yeah, if you have fewer people clamoring for the TOL, you can work slower and charge more. Think Apple. Think Porsche. Then there's less of a need to farm everything overseas and release a new clock speed every month that's .01 GHz faster than the last.

Moore's Law

[DJPenguin]

Dammit, I always thought that came from Mandy Moore, not some old guy!

Bah.

[Jonboy X]

What does Google know? All the do is run a cheesy web site. Heck, they don't even serve up and big graphics...

Add and subtract CPUs like RAM

[Neil Watson]

I've said this before, we need to forget about clock speeds. Multiple CPUs that can be added or subtracted based on your needs is a better way.

For my money

[Bob Abooey]

I would rather see faster hard drives with more throughput and faster memory and faster network connections, etc. It's too bad we haven't seen Moore's law for hard drive I/O. For the common PC person these are most likely the bottlenecks not the chips themselves.

Of course there isn't as much glory in making a hard drive when compared with a CPU with a cool name.

Re:Does Moore's Law actually hold back development

[nolife]

You mean like CD drives?
2x, 4x, 6x, 8x, 10x, 12x, 16x, 20x, 24x, 32x, 36x, 40x, 44x, 48x, 52x, 56x, 72x

How much technological advancement in cd reading was really required for each of these time delayed jumps?

Re:Does Moore's Law actually hold back development

[vidarh]

I think this is highly unlikely. It would make sense if there were no competition in the CPU space, but certainly if AMD had something vastly superior to their current offerings ready for release, they'd release it to get ahead of Intel, and the other way around - the two of them lose enough business to the other that it would more than outweigh any extra revenue they'd get from holding back technology.

Perhaps if both of them have significant technology to hold back and did a deal to hold it back to keep prices up, but in that case they could've just as well agreed to fix prices and there seem to be no indication of that.

Hey

[SlashdotLemming]

Now that we have our Moore's Law story, does that mean the Perl book review is coming soon?

More Expensive is no longer better

[diablobynight]

The problem is that the processors that are out for home use such as the new Barton chip and the newer Pentium 4s are amazing processors. So the speed increase overall is not even close to a power of ten, if it's even close to twice as fast I would be suprised. Especially with NForce motherboards offering dual memory channels and server board like performance. A lot of companies switched from these overated quad Xeon systems and other such expensive servers a couple of years ago, A local company last year, switched from buying overtly expensive servers with raid drives that cost ten times as much than IDE for hard drive space and the quad Xeons that they were only getting 10-15% performance increase from that 4th 560$ processor, and now they run Dual Athlon systems. With standard ata 133 drives inside, two deal with the lack of caching in an IDE system, They installed an impressively low priced IDE SAN sold by IBM and now have 10 times as much hard drive space as they used to for slightly less money. The Dual Athlon Systems, built by me actually (Which I admit I was scared to put in a server environment, because of previous instabilities in ADM chipsets), these systems run superbly, and were inexpensive to build by comparison of 8000$ Dell Servers or $$$$$$$ Sun Servers. I am not going to tell you that a dual athlon is a better server than a Sun, no way, I know Suns are rock solid, but since I could build the Athlons so cheap we built 6 servers to share load, run pearl, run SQL, run their web server, run their domain controller, and I dedicated one system just to Lotus Notes, a program I despise but this company loves. Anyhow the point is, the price was so low, they had me build them 3 extra servers and prepare a server image for them. Currently these servers work to share load of the web server and they supply the company with an added 400GB of file space per server, 4-120GB IBM drives per server and I dedicate 80GB per system to empty space. Just because keeping 20GB free per Hard drive is probably a good idea so that defrag processes can run properly. I don't know if this is an answer for everyone, but for what most server rooms handle, mail, file space for the company, print server, web server, CAD drawing servers, These smaller powerhouses have been doing excellent. I am glad to see companies that no longer believe IBM, Dell, Intel, more expensive is better.

Re:More Expensive is no longer better

[boskone]

Are those public webservers that are also hosting confidential internal data? If so, you may want to rething your strategy. If it is internal web servers taht are pitching data, then fine, but if it's a mix of internal and external functions, I would suggest you may be vulnerable to data theft, tampering or loss.

Re:More Expensive is no longer better

[diablobynight]

ummm...I don't need a lesson in net security. I assure you that the systems residing in the DMZ are properly secured. They are running Redhat 7.3 and I know my securities, the rest of the network is behind a hardware firewall. So I understand your trying to be helpful but it's insulting that you think from what I told you that you could tell I had a possible security problem.

Re:More Expensive is no longer better

[Blimey85]

but it's insulting that you think from what I told you that you could tell I had a possible security problem

Isn't there ALWAYS a possible security problem? I also run RedHat on the servers I manage and there are several security patches released each week, which get installed immediately on my servers. But you have to realize that until those patches are installed, a security problem exists. There may be a new patch tomorrow to fix a problem that exists today.

There is always a security issue somewhere that can be exploited... but from the sound of things, you know your shit and are doing what you can to minimize the risk. All I'm saying is that you should never get over-confident and start to believe that your boxen are secure. Security is an ongoing process, not a destination.

Re:More Expensive is no longer better

[boskone]

My apologies. It sounds like you have things well in hand then.

Best Regards

Missing the point

[raduf]

I've read most of the comments so far and they don't seem to get the point of the article.
The point is that except for a limited class of applications (multimedia, games), most of the things you can do on tommorow's computers you can do on today's. And that it's becoming incresingly expensive to follow Moore's law while it's less and less necessary.

Much more important will probably become the price and maybe other factors (versatility? miniaturisation? power consumption?). Imagine dirt-cheap wireless chips and p-3 like microprocessors. Think about the applications for a moment.
With the right protocols in place it could mean unlimited bandwith anytime anywhere, just for starters.

When the world will be 100% computerized it won't be because of supercomputers. It'll be because of cheap small chips and smart software.

Re:Missing the point

[Un pobre guey]

Lots of people here are missing the point. This article doesn't argue that Moore's law is no longer valid, nor that Google isn't taking advantage of it. It's arguing that the IT industry (especially Intel) has built a business model around the notion that the business world will consume computing resources as quickly as Moore's law provides them.

I think most of us did get that that was the point. Many of the posts are questioning whether it's really true. The Google example has proven to be quite a distraction for those that believe that Eric Schmidt is stating a truth that will invariably come to pass. I don't think he is lying, but he is definitely saying something that usually goes with disclaimers along the lines of

This statement contains forward-looking statements regarding financial and contractual commitments that are subject to risks and uncertainties. These risks and uncertainties could cause actual results to differ materially from those described in such statements. The reader is cautioned not to rely unduly on these forward-looking statements, which are not a guarantee of future or current performance. Such risks and uncertainties include long-term program commitments, the performance of third parties, the sustained performance of current and futures products, financing risks, the ability to integrate and support a complex technology solution involving multiple providers and users, and other risks detailed from time to time in the company's most recent SEC reports, including its reports on From 10-K and Form 10-Q.(boilerplate disclaimer from sgi.com)

In other words, if he gets a good deal on Itaniums or Athlon64s, or if Google decides to keep huge parts of their databases in-memory and therefore could use the larger address space, or if they have to compete head-to-head with someone who only uses 64-bit CPUs, or whatever other reason, they will start buying them immediately.

What has changed is that business no longer has a need to buy the latest and greatest it has to offer, because the uptake of technology has reached the point of diminishing returns. This means that business no longer cares for the most powerful hardware, it no longer gives them a competitive edge in their industry.

Again, only if you believe it. I don't know about you, but my employer (a well-known biotech frim) sucks in as much processing power as it can reasonably get a hold of. Our needs are getting bigger by the day. I am writing this on a dual-CPU Dell desktop with 1 Gb of RAM, and I often am frustrated that I can't load XML files larger than a mere 30 or 40 Mb into memory, or that my 900 Mb log files slow everything to a crawl when I load them into Textpad. My kids play games that are just not as fun unless you have one or more 1.5+ MHz CPUs and at least a GEForce3.

Michael S. Malone's article in Red Herring was just that, a red herring.

One size does not fit all

[mpthompson]

At my company, Atomz, our focus is on a different segment of the search market than Google's -- hosted site search rather than Google's primary emphasis on Internet wide search. Therefore we most certainly have different technology requirements to satisfy our unique business needs. At Atomz we manage search and content management services for more than 50,000 web sites we are constantly examining issues that impact our gross margins.

From this perspective the overhead to maintain our server infrastructure has much more to do with maintaining proper hosting facilities (clean bandwidth, power, backups, etc...) and the cost of IT people to manage and maintain our systems rather than the cost differential between low-end and high-end CPU architectures. For us, fewer systems means lest costly data center real estate and a fewer number of people required to maintain the systems which directly impacts our margins. This tends to push us more towards a fewer number of servers with more powerful CPUs and better I/O performance than a larger number of bottom-end systems would provide us.

We share some of the same co-location facilities with Google some I'm more than a bit familiar with the Google's approach to maintaining their Internet wide search service. In our segment of the search market we could not stay in business using the same approach of maintaining a massive number of very low-end servers. But then again, common sense would indicate this to be so because are in a different segment of the market with it's own unique needs.

The overall point I'm trying to make is that it is a gross oversimplification to assume "What's good for Google is necessarily good for the industry as a whole". The availability of more powerful CPUs and higher performance storage will have a more dramatic impact on some businesses than others and the continued role of Moores law for another decade will be welcomed and embraced by many businesses.

Re:One size does not fit all

[Queuetue]

Nice ad. I certainly hope that posting your company's name brings in the revenue boost you were hoping for.

If I ever need some "Enterprise Web Site Content Management" or some "Site Search Engine Solutions," or even perhaps a website that uses broken javascript to navigate improperly, I'll give you a call.

Moore's Extension?

[Tikiman]

Apparently nobody has noticed the Sony Bono Moore's Law Extension Act, which retroactively extended Moore's Law an additional 10 years after Moore's Law was due to expire

I like your logic

[diablobynight]

But remember that if the Head Node goes down in a cluster, your as well up shit creek. But I agree that a cluster is better. Although space could be a problem, as in physical space, but no worse than huge rack mounted servers.

Re:It's heat. It's power. It's waste.

[Junks Jerzey]

Couldn't it be that the people developing the Itanium recognise the truth of your first statement, and are fixing that problem? In that case, isn't the need for new compilers a good thing?

The Itanium comments have been along the lines of the instruction set being very difficult to optimize, and that current optimization technology is not up to the challenge. That doesn't sound like they've addressed anything.

IT isn't everything.

[Noren]

This arguement is ignoring a major point. Sure, home PCs, web servers, search engines, databases may all get fast enough that further computational speed is irrelevant.

But when computers are used for crunching numbers we still want machines to be as fast as possible. Supercomputers still exist today. Countries and companies are still spending millions to build parallel machines thousands of times faster than home PCs. They're doing this because the current crop of processors is not fast enough for what they want to calculate.

Current computational modeling of the weather, a nuclear explosion, the way a protein folds, a chemical reaction, or any of a large number of other important real-world phenomena is limited by current computational speed. Faster computers will aid these fields tremendously. More power is almost always better in mathematical modeling- I don't expect we'll ever get to the point where we have as much computational power as we want.

Re:IT isn't everything.

[InadequateCamel]

"More power is almost always better in mathematical modeling- I don't expect we'll ever get to the point where we have as much computational power as we want."

If the automotive industry followed this model we would all be driving LeMans 24-hour cars with V16 engines to get to work in the morning, and despite the fact that your 2001 Porsche Boxster works just fine, every time you take it to the garage the mechanics insist it would be much cheaper to haul the engine out and cram at least a V12 in there.

With the exception of research, 99% of computer users have enough power, or rather could have enough power if it were a little cheaper. "Want" is the key word in the above excerpt; I currently run an Athlon 1.2 and GeForce3, but I WANT a faster processor and a better vid card. The point of the article is that people have trouble distinguishing between the two. Google wisely realised that a number of cheap processors is a much better approach than the latest bleeding-edge technology.

At no point did they suggest that research is wrong and has to be stopped ASAP. Their problem is with the way our resources are being allocated: rather than try to produce our current outstanding technology in a more efficient and accessible way we are fixated on making everything faster and bigger.

Re:IT isn't everything.

[Noren]

I guess this is all a matter of perspective. I doubt a race car driver will care if Toyota announces that they don't need to make the engine in a Camry any more powerful in the next model year. He'll still want the best possible engine for his race car and will upgrade it whenever possible.

With the exception of research, 99% of computer users have enough power, or rather could have enough power if it were a little cheaper.

So, for some applications computers are good enough. I think that the 1% of people using 50% of the clock cycles to do research is what computers are for, and the 99% of people surfing the web or playing Quake is a nifty application for computers but beside the main point. (Warning: 86% of the statistics in this paragraph were made up on the spot)

I "want" faster computers so that I can model chemical systems more accurately, not so that I get 300 fps instead of 200 fps playing Quake. Just because you don't have a real need for a faster computer doesn't mean no one does.

On the other hand, the mass market drives R&D and production, so as a practical matter the research dollars for faster processors is likely to taper off if demand does. In this we agree. From my perspective, this turn of events is unfortunate.

Re:IT isn't everything.

[InadequateCamel]

(Warning: 86% of the statistics in this paragraph were made up on the spot)

Hee hee hee

I have used computers for geometry optimization of molecules, and a 20-atom structure can take days on a high enough basis set. Just last summer I was calculating electron density around a moderately-large molecule using HyperChem, a Windows-based PC molecular modeling program, but I ran out of space on my biggest HD partition (10 of 40GB) because it has to stockpile all it's solutions, and I was still only 0.1% finished after 1 day of computation on my Athlon 1.2 with 0.5GB RAM. So yes, I completely agree with you about needing affordable faster/stronger computing.

Unfortunately, except for these applications, there is no need for faster computers in the mass market, but _only_ want. If it was announced tomorrow that we are at the top of the x86 performance mountain, I figure that 8/10 of users would get along just fine with their current processor, and the rest would be able to get a very cheap upgrade to serve the purpose.

So I share your perspective, but I am taking into consideration the way things are in the mass market now and how things should be, ie. the majority of people and businesses should perhaps be more concerned with "How can I get the most out of what I have" rather than "How can I find a way to buy that new CPU".

Other implications of Moore's law

[hey!]

Moore's law not quite what most people think. If I'm not mistaken, it isn't that processor power will double every eighteen months, but that transsistor density will double. Processsor speed doubling is a side effect of this.

I think there will always be a market for the fastest chips possible. However, there are other ways for this trend to take us rather than powerful CPU chips. These would include lower power, lower size, higher system integration, and lower cost.

The EPIA series mini-ITX boards are an example of this. Once the VIA processors get powerful enough to decode DVDs well, it is very likely that they won't need to get more powerful for most consumer applications. However, if you look at the board itself (e.g. here),
you'll see that component count is stil pretty high; power consumption, while small, still requires a substantial power supply in the case or a large brick.

When something like this can be put together, capable of DVD decoding, having no external parts other than memory (and maybe not even that), and the whole thing runs on two AAA batteries, then you'd really have something. Stir bluetooth (or more likely its sucessors) into the mix and you have ubiquitous computing, capable of adapting to their environement and adapting the environment to suit human needs.

Re:Other implications of Moore's law

[InadequateCamel]

No, he altered the definition once transistor density growth slowed a little. Moore's Law v1.1 relates to data density instead, so performance is now a little closer to the truth.

Re:Other implications of Moore's law

[skeedlelee]

I'm about to rant a little here.

IIRC, there an article a while back (discussed here) that reviewed Moore's law, as Moore used it over a number of years and found that Moore himself seemed to redefine it every couple years. It's a marketing term which describes the general phenomenon of faster computers getting cheaper in a regular way.

You're right though, Moore never really talked about doubling of 'processor power,' he discussed things in terms of devices such as transistors. Trouble is sometimes RAM was included in the 'device' total sometimes not... it's easy to fudge a bit during the slow downs and speed ups if you change how the thing is defined.

Top it off with the fact that the whole thing was eventually cast in terms of the cost optimal solution. Given the degree to which the size of the market for computers has changed I'd say that this is a very difficult thing to define. As everyone is likely to point out, commodity desktop PC's have a very different optimum from massive single-system image computers. Of course, if you consider that a calculator is a computer, be they $1 cheapo's or the latest graphing programable whoopdeedoo they are all computers. There are so many markets for computers now, each with their own optimum that it's pretty artificial to talk about Moore's law at all. I've never seen anyone plot out Moore's law with a bunch of branches. Further, cost optimal becomes pretty subjective in all the markets when there are so many variables. Finally, there are points where Moore's law breaks down... the number of devices in cheapo calculators probably hasn't changed much in the last few years, but the price changes. Moore's law doesn't really allow for this sort of behavior, that there is a maximum necessary power for a certain kind of device, if it doesn't have to do anything else, then the complexity levels off and the price goes down. This may well happen at some point in the commodity sector. It is possible that the number of features in a conventional desktop will level off at some point. Hell with a $200 WalMart cheapo PC, maybe we're there now...

Intuitively, everyone applies Moore's law to desktops but there's no particular reason to do so. Considering the history of it the massive mainframe style computer is probably the best application of it, but this is seldom done. Mainframes these days can be a complex as you're willing to pay for, which pretty much means that there is a cost optimal solution for an given problem, not just for fabrication, which is what Moore was talking about. Seems like we have turned a corner, it's time to redefine Moore's law yet again.

Since when...

[Chocolate Teapot]

...did Moore's opinion become law?

Organic computers...

[danro]

Great idea. Make a computer that sees humans as a parts/food source. Add a delicious incentive for robot underlings to revolt.

No way!
My box will be a vegetarian just like me.
...or, will it. Why did the cdrom just start spinning for no reason...

Moore's Law of dot.com bubble

[peter303]

The Moore's Law hype resembles market analysts' prediction why the DJIA could go to 36,000 and beyond. Internet companies did not need profits to go public; they even didnt need revenues. Of course, the InterNet will change a lor of things, but not that fast. And Moore's Law will continue to some degree for some time.

how does google's comment violate moore's law?

[k2enemy]

moore's law says nothing about the speed or power of a chip, only the density of transistors. if you hold the size of the chip constant, the number of transistors can increase, or if you hold the number of transistors constant, the size of the chip can decrease.

if google uses a large number of low speed chips, they will still benefit from smaller sized chips with lower power consumption and lower price.

He maintained an emacs fork; he was no slouch.

[emil]

While you may validly question his business acumen, he has worked with RMS, JWZ, and knows everybody. He is a reasonable coder and a team player; we need more of him.

This has happened before!

[HiThere]

The mainframes basically stopped at 32 bits. There were models that went to 128 bits, and CDC liked 60 bits, but the workhorse (IBM 360, etc.) never went beyond 32 bits.

Perhaps the next step instead of being towards larger computers will be towards smaller ones. Moore's law remains just as important, but the application changes. Instead of building faster computers, you build smaller, cheaper ones. The desktops will remain important for decades as the repository of printers, large hard disks, etc. And the palmtops/wristtops/fingernailTops/embedded will communicate with them for archiving, etc.

This means that networking is becoming more important. This means that clusters need to be more integrated. I conceive of future powerful computers as a net of nets, and at the bottom of each net is a tightly integrated cluser of cpus, each more powerful than the current crop. These are going to need a lot of on-chip ram, and ram attached caches, because their access to large ram will be slow, and mediated through gatekeepers. There will probably be multiported ram whiteboards, where multiple cpus can share their current thoughts, etc.

For this scenario to work, computers will need to be able to take their programs in a sort of pseudo-code, and re-write it into a parallelized form. There will, of course, be frequent bottle necks, etc. So there will be lots of wasted cycles, but some of them can be used on other processes with lower priority. And at least each cluster will have one cpu that spends most of it's time scheduling. etc.

Consider the ratio between gray matter and white matter. I've been told that most of the computation is done in the gray matter, and the white matter acts as a communications link. This may not be true (it was an old source), but it is a good model of the problem. So to make this work, the individual processors need to get smaller and cheaper. But that's one of the choices that Moore's law offers!

So this is, in fact, an encouraging trend. But it does mean that the high end cpus will tend to be short-term solutions to problems, faster at any particular scale of the technology, but too expensive for most problems, and not developing fast enough to stay ahead of their smaller brethern. Because they are too expensive to be used in a wasteful manner.

Perhaps the "final" generation will implement these longer word length cpus, at least in places. And it would clearly use specialized hardware for the signal switchers, just as the video cards use specialized hardware, though they didn't at first. But the first versions will be built with cheap components, and the specialized hardware will only come along later, after the designs have stabilized.

Proposition

[Anonymous Coward]

Instead of indulging in this fetish over GHz and how many transistors can fit into a thumbnail of silicon, why don't we instead say, "OK, the hardware guys have that part of the picture taken care of. All the software guys should now work on making programs as secure, robust, well documented, and usable as possible"?

I know the answer to that, of course--it's not nearly as sexy to to that kind of work as it is to overclock your new CPU or hack code (as opposed to doing real software development). But think of what computers could be if the software guys improved their work at even half the rate the hardware guys do...

Re:Proposition

[Un pobre guey]

But think of what computers could be if the software guys improved their work at even half the rate the hardware guys do...

What makes you think that isn't occurring? How do you know software isn't being improved at a vastly greater rate than hardware? Do you have a metric, or any credible evidence either way? Surely you're not just hopping on some banal blah blah blah bandwagon?

Look around you and tell me it could all have been designed, manufactured, traded, transported, resold, deployed, powered, and used by the average Joe with software from 10 years ago.

Honestly, remind me what I pay you people for
Dr. Evil

number of transistors on a chip

[Anonymous Coward]

Anybody know how close is it to being impossible to increase the number of transistors on a chip, due to physics, i.e. restrictions via tinyness and being unable to transmit electrons etc.

Article Full of Overblown Rhetoric

[lamz]

This article is full of overblown rhetoric. It goofily applies Moore's Law to too many other things, like Dot-coms. Note that at no point in the article is Moore's Law clearly stated -- it would spoil too many of the article's conclusions.

That said, I remember the first time I noticed that technology was 'good enough,' and didn't need to double ever again: with the introduction of CDs, and later, CD-quality sound cards. Most people are not physically capable of hearing improvements if the sampling rate of CDs is increased, so we don't need to bother. Certainly, people tried, and the home theatre style multi-channel stuff is an improvement over plain stereo CDs, but it is an insignificant improvement when compared to CDs over older mono formats. Similarily, the latest SoundBlaster cards represent an insignificant improvement over the early beeps of computers and video games. (Dogs and dolphins might wish that audio reproduction was improved, but they don't have credit cards.)

Back in the early 80s, when most bulletin board access was by 300 baud modem, paging of long messages was optional, since most people can read that fast. Of course, we need faster modems for longer files and applications, but as soon as say, HD-quality video and sound can be streamed at real-time speeds, then bandwidth will be 'enough.'

The Real Problem - Computers Are Too Slow

[PingPongBoy]

Seriously at this point most people don't need 1Thz CPU's

We are at a point where the next GHz might not mean a lot but the reason is not computers are too fast. The problem is that computers are still too slow.

I recently bought Dragon Naturally Speaking thinking that my 2GHz Pentium 4, which does everything so nicely, would make speech recognition very easy. The software cranks up the CPU but it still can't a lot of what I say. It lags and requires a lot of training. Its technology is based on understanding strings rather than individual syllables, so I can train it with special phrases, but it misses a lot of new phrases.

We need very fast computers that are very small. Applications:
- recognition (speech, facial, etc.)
- automatic car driver
- natural language understanding

I want to carry the computer perhaps on a wristband and have it multitask.

Re:The Real Problem - Computers Are Too Slow

[MikeFM]

I think AI and ultra-realistic VR are things that no matter what speed you throw at it aren't going to crack easily. These are parallel processes and they need massive grids of processing to work well. For real improvements I think you'll have to wait until nanotech computing is a reality. Trillions of tiny computers working together can do what one big fast computer can't.

Also there is a long way to go in software design in these kinds of fields. More horsepower helps (a lot) but it certainly won't make smarter computers without better software.

Moore's Law Indicates Exponential Intelligence

[PingPongBoy]

It's a law of economics or human behavior. It's more a statistical phenomenon of wavering reliability.

Still, Moore's Law may be a measure of intelligence. If you look at the accomplishments of dumb animals, they don't seem to get anywhere from one generation to the next if you ignore evolution, which is more an indication of being driven by the environment rather than being a master of it.

So far Moore's Law says we can double about every two years. How intelligent is that?? One answer may be "exponential".

The Turing Test looks for the presence of intelligence. Moore's Law measures the degree of intelligence, civilization, resourcefulness, etc.

Disruptive technologies

[bstadil]

What you are describing is more a case of Disruptive Technologies than Moore's law per se.

Obviously Moore's law is fuelling some of this, but it is not the only reason. Second Moore's law is about integration as well as performance. The former is much more disruptive than the latter.

AI

[Un pobre guey]

There are on the order of 10^15 synapses in the human brain. Let's pretend for a moment that there is a rough equivalence between a logic gate or a memory location in a computer and a synapse in the human brain. Let's further pretend that someone may have the intention of constructing a computer with roughly the computing power of a human brain

We may continue this reckless pretending (orgy of pretense?) by supposing that such a person might want that computer to be of a reasonable size, perhaps even capable of sitting on a desktop. Other people, unsatisfied with any and everything, might insist that the machine be even more powerful than a human brain. Much more.

Now I ask you, will such a machine be made of 32-bit CPUs, or even of 64-bit CPUs? Will it even be made of CPUs, or even have any silicon in it at all? Can it be built if we stop the "runaway train, roaring down a path to disaster, picking up speed at every turn, and we are now going faster than human beings can endure?" I'll save you the trouble of activating a few billion synapses: no, it will not. Far from it.

Moore's Law (or Moore's Ex-Recto Conjecture, as one poster implied), will go on as long as people can make a few bucks making it go on. And of course, its path will be littered with jewels, chaos, marvels, disaster, fetid waste, and vast new possibilities we can only dream of.

This, as usual, in spite of loud, hypey articles in "lifestyle" magazines.

Evil Woz? BAH!

[Anonymous Coward]

Steve Wozniak could never be evil, even if you cloned him. He'd just be a lot of fun to hang out with. WAAHH! I want my own Woz clone!!!

Real-Time Cross Platform Video Processing

[RobPiano]

I am currently doing real-time video processing for a digital musical instrument (video tracking of a performer). The video is processed on a PC with mediocre specifications and the audio is processed via OSC on a decent G4 on the other side of the room.

Could I use a faster computer to do this? Yeah I suppose so, but really speed and bit size don't matter much. Hell, I don't get true 24 bit audio using a 24 bit card. What I need is quieter, STABLE machines that communicate with each other quickly. I need hard disks that do not crash (already lost one on a brand new G4), I need gigabit that is really gigabit, and a slew of other practicle things. None of them involve a faster processor or 32 more bits.

Computers will get faster yes, but really the MHz battle is over in my mind. I need all the other technology to become more stable and the less glamourous bottle necks to be overcome.

Soon!

-Rob

Mini-ITX power supply of doom over here...

[MsGeek]

Not exactly two AAA batteries, but small and sufficiently powerful to run 512MB of RAM and two 120GB hard drives: http://www.mini-box.com/PW-60.htm.

This works with all iterations of the EPIA platform and will happily use the power supply that comes with a Cubid (common mini-ITX factor case about the size of a 1" 3 ring binder) case. No need for a fan, so it's totally silent.

Re:Mini-ITX power supply of doom over here...

[MsGeek]

Uh...I mean that it will use the transformer brick that comes with a Cubid. It replaces the power supply board for the Cubid.#^_^#

That article is crap

[Animats]

I read that article, and was underwhelmed.

There's general agreement that the technology on which Moore's Law is based, optical lithography on flat silicon, hits a wall about a decade out. That's when atoms are just too big. There's argument over whether some other limit is reached first. The cost of fabs continues to increase, while demand is flat. Economics may present a limit before atomic structure does. Power consumption per unit area is getting out of hand. But the industry has overcome such problems before. Atom size, though, is fundamental. When we hit that, a new idea is needed.

There's no replacement technology in place to carry on from traditional silicon wafers. Many alternatives have been proposed, but they're all more expensive. Building up 3D chips one layer at a time isn't cheaper than what we have now, and the cooling problems are worse. Nanotechnology, biocomputing, optical computing, and quantum computing are nowhere near ready to take over. Their time may come, but it's not close. Each of those technologies has more than a decade of development behind it, without products emerging. That's a bad sign.

Consider, by comparison, turbine engines for ground vehicles. Clearly they are possible; turbine-powered cars, trucks, and buses have been built. But they're not cost-effective. Most of the replacements for flat silicon look that way.

As for the Inanium, price/performance is clearly lower than with IA-32, so it's no surprise that people with big server farms like Google have rejected it. But that's architecture, not fabrication, and has nothing to do with Moore's Law.

I guess in this case I just wasn't concerned

[diablobynight]

Ummm...if someone really wants to hack into a company that makes coffee mugs and sells them in bulk, they have way too much time on their hands. lol

The Dilemma at work perhaps?

[chaeron]

I wonder if this isn't a symptom of the Innovator's Dilemma at work?

Might presage a bit of an upset of the "big boys of silicon" (eg. Intel, AMD, et al) if it is.

The starlets in Hollywood will be pleased if that happens. More for them.

But what if Moores law is too slow?

[argoff]

Everyone seems to be acting like Moore's law is too fast, that over the next centruy our technology could never grow as fast as it predicts. However, consider for a moment that perhaps it's too slow, that technology can and will grow faster than it's predictions like it or not. Yes silicon has limits, but physics wise - there is no law I know of inherent in the universe that says mathematical calculations can never be calculated faster than xyz, or the rate of growth in calculation ability can never accellerate faster than abc. These constraints are defined by human limits, not physical ones.

In fact, it could be argued that Moore's law is slowing down progress because inverstors see any technology that grows faster than it predicts as too good to be true, and therefore too risky to invest in. However, from time to time when companies have been in dire straights to outdo their competitors "magical" things have happened that seem to have surpassed it for at least brief periods. Also, from what I understand, the rate of growth in optical technology *IS* faster than moores law, but people expect it to fizzle off when it reaches the abilities of silicon - I doubt it.

The last time Intel was declaring the death of Moores law was when they were under heavy attack from predictions that they couldn't match advances in RISC technology. Funny, when they finally redesigned their CPU with RISC underpinnings - these death predictions silently faded away. (at least till now) I wonder what's holding them back this time?

Re:But what if Moores law is too slow?

[nobbis]

Physics does constrain the speed of computation. Read some David Deutsch (such as The Fabric of Reality) for the low-down.

Re:But what if Moores law is too slow?

[jbischof]

I read an article a while back about the theoretical limits of computers.

Contrary to what some people think, quantum physics and thermodynamics define the limits of what a computer can do.

For example, you can't send data faster than the speed of light, and you can't have two memory blocks closer together than one plank's constant. Likewise you cannot store more information than the total amount of information in the universe etc. etc.

According to physics as it is today, there are dead ends for computers where they cannot get any faster, bigger, or more powerfull. We may never reach those limits, but they still exist.

Re:But what if Moores law is too slow?

[Anonymous Coward]

predictions about the death of Moore's law never faded away, every month there is a new person talking about the death of Moore's Law and that has been going on since he first postulated it.

Re:But what if Moores law is too slow?

[oconnorcjo]

predictions about the death of Moore's law never faded away, every month there is a new person talking about the death of Moore's Law and that has been going on since he first postulated it.

I never said "Moores law" was over and I remember having a 386 that ran at 33 mhz. Now that computers can run a thousand times faster, I am say that it is probably true. When the 486 was around computers did not even need a heatsink. Now CPU's need a fan and a heat sink or they will fry in less than a minute. 0.13 um is the latest and greatest in silicon shrinkage and even if CPU's shrink to 0.09 um (which they might in six to ten years), the purity and the precision needed to print a chip today is getting to the extremes of what manufacturing is capable of in mass production. "Moores law" has been slowing down over the last 25 years. It used to be it doubles in 6-9 months, then a year and now it is 18 months. Things might start doubling every 36 months and then fade away altogether. I am not saying we won't see ten to twelve ghz CPU's (we will) but that 30 ghz won't be seen in my lifetime (except maybe in some research lab to prove it could be done). If things start "only" doubling every 3 years then that would mean we will have 48 ghz CPU's in 24 years. Heat has been an issue for the last ten to fifteen years and it has not gone away! It has been fought by adding bigger heatsinks and fans on top of the CPU combined with shrinking the "die chanels" inside the CPU to give off less heat. There is a limit to how far "the chanels" can be shrunk and how big the heatsink/fan can be. AMD and Intel have both spent a Fortune (billions) to go over to .13 um and I can't see them going over to .09 anytime soon (can you?). The heatsink/fan combo is already huge so water cooling could be next but I think it is close to the end of the line.

Re:But what if Moores law is too slow?

[oconnorcjo]

If things start "only" doubling every 3 years then that would mean we will have 48 ghz CPU's in 24 years.

Whoops ... I doubled the 12 years. That should be 15 years (but then again, I doubt we will have 48 ghz in 50 years).

Re:But what if Moores law is too slow?

[Newander]

Correct me if I'm wrong, but the article is not about Moore's law failing to predict technological advance, but rather that if the industry continues in the same way it has been, it will destroy itself with ever decreasing profit margins. That you can make a faster computer, but who's going to buy them every 18 months. I know I can't afford a new computer that often. Hell, I just got rid of my PPro 166.

Re:But what if Moores law is too slow?

[Hognoxious]

I never said "Moores law" was over and I remember having a 386 that ran at 33 mhz.

Luxury! I used to dreeeeaaaam of having a 386.

A supplement to Moore's Law: ... and the proportion of computing power employed for useful purposes halves every 6 months.

Moore's Law of /. Articles about Moore

[Anonymous Coward]

It has only been a month since the last /. troll fest about Moore's law. Are these fiascoes going to become exponentially more frequent until they consume all of the available bandwidth in the universe? I say we should put a lameness filter limit on them. Say No more than once a year?

Google's fast because...

[unsung]

All of this talk about faster processors - don't forget that Google is also fast (for the user) in part because their website isn't choked up with unnecessary graphics and scripts. A fast processor isn't going to expand the network bandwidth.

Another reason that they're fast is because of the relevancy of their hits. Typically I don't have to trudge through 10 pages of hits to reach my search term. Again, raw processing power doesn't have much to do with this.

Re:It's True!!!

[Anonymous Coward]

i was rather worried that the link would lead to goatse.cx. i guess i'm glad that it didn't, but it would have been funny.

Don't miss this related article...

[Dave21212]

Recently (Jan 06) on /. there was a well written if not rather technical essay on Moore's. For those of you who want the real scoop...

Moore's Law Disputed

The Lives and Death of Moore's Law by Ilkka Tuomi

Moore's Law has been an important benchmark for developments in microelectronics and information processing for over three decades. During this time, its applications and interpretations have proliferated and expanded, often far beyond the validity of the original assumptions made by Moore. Technical considerations of optimal chip manufacturing costs have been expanded to processor performance, economics of computing, and social development. It is therefore useful to review the various interpretations of Moore's Law and empirical evidence that could support them.

Such an analysis reveals that semiconductor technology has evolved during the last four decades under very special economic conditions. In particular, the rapid development of microelectronics implies that economic and social demand has played a limited role in this industry. Contrary to popular claims, it appears that the common versions of Moore's Law have not been valid during the last decades. As semiconductors are becoming important in economy and society, Moore's Law is now becoming an increasingly misleading predictor of future developments.

Moore's Law

[Anonymous Coward]

Truthfully Moore's law is merely a theory waiting to be disproven. As all laws exist this way. The law of gravity. The law of magnetism. The laws of physics. While they seem to work perfectly now. What about that rare opportunity that one might be disproven. Utter chaos my brothers.

Truthfully I don't really care what Google is going to do. They provide me with a service that works quite well. I have never complained except for every search about Real and Nude and Photots and Britney Spears. I always get fake ones. Sad.

I think chip companies will resolve something over time. So long as I can continue to play my EQ or WC3 or whatever else I do on my computer with relavtive speed and ease then I am happy. Well that and browse for porn. Mmmm. Faye Valentine.

It's ignoring Moore's law that's dangerous.

[AnotherBlackHat]

Michael S. Malone says we should forget Moore's law, not because it isn't true, but mainly because it has become dangerous.

How someone can look at such a devistating example of Moore's law in action and conclude that it's the law that's the problem is beyond me.

It's ignoring Moore's law that's dangerous.
If Moore's law continues (and it shows no signs of stopping, even after more than 35 years) then eventually will all have the computing power that Google has now, sitting on our desktops. The implication is obvious - in another decade, Google will fail. Not because their database is inferior, but simply because we'll all have our own copy of it.

To put it another way, most of us have fixed resource requirements. Once we can create full ray-traced 3d images at 72FPS, we're done. Most of us won't need more powerful computers. (We'll want them of course, but we won't be willing to pay for them.) If you think the dot-bomb has gone off, think again. Sales of high-end computers are already declining, and it's only going to get worse, all because of Moore's law. Forget Moore's law? Might as well take cyanide, it would be quicker.

-- this is not a .sig

Example of problem

[Anonymous Coward]

I'm not disputing that they exist. But I'm drawing a blank. Can someone please give an example of a computing task that CANNOT be subdivided into smaller tasks and run in parallel on many processing elements? The kind of task that requires an ever faster single processor.

The only short description one I know - matrix inversion for a general square matrix. All data values depend on all solution values. Special matrix forms may work better parallel but not the general case

Things that work well in parallel are things that have nicely defined chunks of the answer that don't require any of the rest of the solution.

The Google problem works nicely in parallel. Each of thousands of computers knows only one page of a dictionary. You ask them all at once and only the ones with the answer responds.

Raytracing also works well in parallel. Each pixel doesn't rely heavily on any other pixel value. Each pixel is expensive to compute and is cheep to transmit.

Price isn't the only problem

[roystgnr]

If you could get THOSE down to $1 each so they could be used in embedded apps from clothing to toasters you would be giving engineers, designers, and inventors a lot to work with.

I suspect that even if today's 1 Ghz CPUs could be produced for $1 each, they would be much more practical to use in toasters than in clothing. Well, except maybe legwarmers.

Confusing 64-bit with faster processors

[Anonymous Coward]

IIRC, Google only stated that they would not use Itanic processors: ie. they see no need for 64-bit computing.

They did not state that they would not want faster, better 32-bit processors.

There seems to be a general mis-conception that 64-bit processing is somehow faster than 32-bit processing. In many cases it is not.

In my field, vendors supply both 32-bit and 64-bit versions of applications for SPARC processors. Why? Because the same program running in 32-bit mode is significantly faster. It is ONLY if you exceed the 4GB per PROCESS limit that there is any benefit in going to 64-bit.

Now, remeber also that 32-bit processors can use more than 4GB of memory. It is just that 32-bit OS-es can not provide more than 4GB of memory to a single process. What processes are google running that might conceivably require more than 4GB PER PROCESS? None. Hence the desire to stay with 32-bit computing!

Missing the point

[Anonymous Coward]

Lots of people here are missing the point. This article doesn't argue that Moore's law is no longer valid, nor that Google isn't taking advantage of it. It's arguing that the IT industry (especially Intel) has built a business model around the notion that the business world will consume computing resources as quickly as Moore's law provides them.

In the case of Intel, their whole business model collapses if they can't convince businesses to buy their new processors every 18 months. It doesn't matter if businesses still buy outdated cheaper processors. It is the bleeding edge processors' revenue that pays for the massive amounts spent on R&D. If they can't sell enough bleeding edge processors, Intel is toast. Associated with Intel is Microsoft, who also depends on businesses buying enough bleeding edge tech in order to force them to upgrade. Associated with both Microsoft and Intel are the myriad of tech companies that have built their business model on the notion that the business world will consume tech as fast as Moore's law can deliver it.

It doesn't matter that Google takes advantage of Moore's law, of course it does. It won't matter that it will eventually buy Itaniums once they become cheap. The point is that they won't do so immediately - at least until the prices come down. This means that Intel's business model is now broken.

If it were up to science alone, Moore's law would continue to hold true. What has changed is that business no longer has a need to buy the latest and greatest it has to offer, because the uptake of technology has reached the point of diminishing returns. This means that business no longer cares for the most powerful hardware, it no longer gives them a competitive edge in their industry.

This is a tremendously significant event, as the article states, and will transform the nature of the tech industry, probably from an R&D and bleeding edge driven one to a commodity driven one. R&D and bleeding edge tech will be relegated to niches. The consequence of this is mind boggling - it could well mean that the industry does not advance as quickly as it could, because business could become a bottleneck.

Since tech progress depends as much on economics and business as science, I'll leave the obvious conclusion to you.

Re:Moot point

[Bastian]

I don't think the issue being raised about Moore's law threatens big iron because companies that produce massive mainframes and supercomputers have been doing what Google just opted to do for years now.

What they seem to be talking about is that Google has realized that Moore's law doesn't just mean that you can get twice as much for the same price in 18 months. It also means you can get the same amount for an ever-decreasing price.

This is a big deal. As IT departments start realizing that if they already have a job that's being handled well by a cluster of, say, two servers that cost $10,000 apiece a year and a half ago but now need an upgrade, they have a few options.

First, they can spend $10,000 on the latest computer that is twice as fast, add it to the cluster, and keep doing that every so often. This is going to start causing some TCO problems with mixed hardware and all that.

Second, they can spend $20,000 on a pair of the latest thing. That might mean siphoning off cash by getting rid of the old machines.

Third, they can buy another of the original machine for $5,000 (probably less). They got what they need, they spent less money than they would with the other two options, and don't have to deal with mixed hardware headaches and all of that. Yeah, the old computers are going to die before the new one does, but by the time that happens it will be even more dirt cheap to replace them with identical machines. Only loss is that as this keeps happening the maintenance costs involved in managing an ever increasing number of machines start to go up.

But Google seems to be handling that problem well, so it may well prove to be not so much of a problem after all.

Google is taking the third option. If other companies take Google's lead, the semiconductor industry, with its rabid policy of pushing forward as quickly as possible and spending enough money to buy Tibet from China each year, it's going to be fucked.

Really, when you think about it, companies that sell Big Iron machines have known this for a long time. Have they been making their computers so damn powerful by fitting them with CPU's the size of floppy disks? No, they've been designing excellent backplanes and buses that allow for the creation of machines with thousands of processors that all access the same memory pool with high bandwidth and low latency.

The decision seems to be working extremely well for Big Iron. It looks like the right option for Google. Now we just have to wait for the rest of the IT industry to catch up.

Not every computer is in a render farm

[drouse]

I think you folks are pulling out some nice edge cases, but most businesses don't run render farms or weather simulations.

There is the electricity:
Even five years from now a 4 GHz, 64 bit CPU will still take more electricity than a 500 MHz, 32 bit CPU. I would imagine that electricity is a big part of Google's continuing expenses, if they can keep that down...

And the upgrades:
We run accounting, billing etc. for a 100 person business on an Ultra 1 (umm ... 1996?). The machine seems a little slow so we will replace it with something, say, 3 years old. After we do that, we might *never* need anything faster. I know, we've said that before -- but we have really reached a point where 3 years ago was enough. For Intel business desktops we have reached and gone past "enough" a while back as well.

What the article is saying is that more and more business are getting to the point that they don't need to upgrade again. And if Intel stops seeing rapid upgrades then it becomes more difficult to keep up R&D.

I think the article overstates thing a bit, but the facts look pretty solid. If you say "the software is going to require more CPU" then I'll reply that software growth is artificial and managed to require hardware upgrades.

If the gamers want to keep upgrading, fine. But business buyers (such as myself) are starting to think about abandoning the upgrade cycle.

All of this, by the way, should be good for Linux. Businesses might start to be interested in an OS that doesn't push hardware into obsolesense every year.

Anyway...

distributed/clustered computing

[guanno]

I think the real application of denser and faster CPUs will be scientific study. In Canada we have a serious deficite of super computer access for research projects, so the academia is increasingly farming out stuff like protein sequencing to shared clusters and remote processing applications similar to GENOME@HOME. However should the 64-bit CPU sucker in enough average home PC users, the price will stay low, which will likely create a tremendous opportunity for advancing affordable scientific research in slightly less affluent nations like my own.

Um, no

[Goonie]

I'm no expert on parallel algorithms, but as I understand it, faster processors aren't going to make it any easier to parallelize problems. In fact, it may even reduce the gains from doing so as the communication overhead becomes a greater and greater part of the cost of the computation on your parallel machine.

Re:Moot point

[Newander]

Great analysis, I couldn't have put my feelings better.

R&D, GDP, revolutions

[NFW]

Seems to me that the most interesting things in the article is the bit about R&D costs following Moore's Law and approaching global GDP (er, GGP?).

I don't care if Google doesn't want faster machines... That just means that Google's application doesn't need more speed, and isn't expected to get more demanding soon.

Other applications - gaming, for one - can make use of faster chips, and will make use of them as quicky as they become available. Are there any game developers and/or artists who aren't wishing they could use 10x more polygons, run collision detection routines 10x faster, do physics with 10x more realism?

So Google is going the way of the word processor - an application that no longer pushes the limits of present technology. It's stupid to extrapolate this into a proclamation that Moore's Law is finished.

But the R&D claim, now that's interesting. How much money is there in other (non-Google, non-word processing) applications? Enough to fund R&D to keep processor speed increasing? For how long? Will the GGP increase enough to support such applications to in turn support such R&D?

More interestingly, will a shift from semiconductor lithography to the next big thing (nanotech, optical computing, _________) allow processor speeds to keep increasing without pushing up R&D? I'm told that Moore's Law fits the data going back into the days of vacuum tubes, relays before that, and mechanical computing before that.

I strongly suspect that the news of Moore's Law's demise is greatly exaggerated.

Fear not!

[Scott Carnahan]

The danger is that soon enough an Intel processor will get hot enough to trigger a fusion reaction in atmospheric hydrogen, turning Earth into a small star.

Fortunately, Intel is working on this as we speak! Here is a Register article with VP Gelsinger's predictions. Here is a nifty photo of an overclocking experiment gone awry.

If that is true...

[jotaeleemeese]

... then the IT industry is a bunch of derided crazy fools. No wonder they are doing worst than other industries.

To base your business planning in an assumption (what Mr Moore said is not a law) is one of the riskier business propotitions I can think of.

Some companies...

[jotaeleemeese]

.... have not realized this and keep using their old business practices.

Sometimes common sense is the less common of all senses.

So stop

[Anonymous Coward]

[Have you tried X recently?] Yes, yes I have. and it should be lightning fast with zero bogging or delays on a P-III 866.. let alone a dual P-III system.. Gnome + Nautilus is slow as a dog. ... And another example is Mozilla and Phoenix.

So you're not actually complaining about X. You're complaining you're running a bunch of other programs on top of X, and they're slow.

Everything you have mentioned SHOULD be seperate optimized apps.

On my system, they are. If you don't like Gnome, don't run Gnome. That doesn't stop you from running separate apps.

Programmers are adding in crap for the sake of adding it. and it needs to stop.

Sounds like users are using crap for the sake of using it. Why don't you stop?

Google on .NET?

[Anonymous Coward]

Here's a good Q&A with Google about Microsoft.

AMD, Hypertransport, Intel, Moore's Law

[abhisarda]

Forget Moore's Law or rather Dumb rule of thumb.

While processor speed has grown fast during the past 6 years, memory speed(bus speed) has grown much slower. Infact, only now is it being pursued seriously.

Once my computer guy told me that having a 400 Mhz computer and PC100 RAM would transfer data only as fast as the PC100 could process it and transfer it. When you build a system and test it with PC2100 DDR and then PC3200 DDR, the difference is much pronounced.

Hector Ruiz is trying to shift attention from Moore's law because in a few years, to get a substantial increase in processor speed, you will need a lot more money than needed today and heat is also a major factor.

You are already seeing a shift in Intel(the proponent of Moore's law) to lesser heat producing processors even while giving better performance than the P-4's. AMD while increasing its processor performance wants data to travel faster so that memory bottlenecks are removed(hence hypertransport consortium).

The performance spotlight is shifting to the memory market because the cost of making data travel faster and thereby increasing overall computer speed is less than increasing processor speed to achieve the same bump in computer performance.

Last Post!

[alpg]

"I said I hope it is a good party," said Galder, loudly.
"AT THE MOMENT IT IS," said Death levelly. "I THINK IT MIGHT GO
DOWNHILL VERY QUICKLY AT MIDNIGHT."
"Why?"
"THAT'S WHEN THEY THINK I'LL BE TAKING MY MASK OFF."
-- Terry Pratchett, "The Light Fantastic"

- this post brought to you by the Automated Last Post Generator...