Slashdot Mirror
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."
BBC Article on the same story here.
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.
Trolling using another account since 2005.
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.
I want my quad 64GHz processor! I want it in 2 years time and I want quad-128Ghz ready by the following year!!!
"I kill you! You no good 56'ing!"
This makes me feel a lot less like a cantankerous, cheap old fart for not replacing my Athlon 650.
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 */
"Every jumbled pile of person has a thinking part that wonders what the part that isn't thinking isn't thinking of"-TMBG
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.
*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
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.
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.
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.
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.
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.
- -
Are you an SF Fan? Are you a Tru-Fan?
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.
How can Moore's law become dangerious?
If you break it, will you explode into billions of particles?
Avantslash - View Slashdot cleanly on your mobile phone.
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.
"Semper in excretum set alta variant"
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.
There are some odd things afoot now, in the Villa Straylight.
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.
Whale
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.)
Do not look at laser with remaining good eye.
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
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
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.
At what price learning? At what cost wisdom? The price is a man's peace of mind, and the cost is his life.
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.
Who said Freedom was Fair?
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!"
The price of freedom is eternal litigation.
From the article:
He gave the Monday keynote at the "Hot Chips" conference at Stanford last August.
There is an abstract of his keynote.
No electrons were harmed creating this post, though some may have been subjected to electrical and/or magnetic fields.
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.
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 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.
Post may contain irony: discontinue use if experiencing mood swings, nausea or elevated blood pressure.
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.
I think we've pushed this "anyone can grow up to be president" thing too far.
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?