Where's My 10 Ghz PC?

An anonymous reader writes "Based on decades of growth in CPU speeds, Santa was supposed to drop off my 10 Ghz PC a few weeks back, but all I got was this lousy 2 Ghz dual processor box -- like it's still 2001...oh please! Dr. Dobbs says the free ride is over, and we now have to come up with some concurrency, but all I have is dollars... What gives?"

Comments

Asymptotic

[dsginter]

We've found the limits of silicon and hard drives and they are being approached asyptotically. Relax...

Re:Asymptotic

[BrianHursey]

True we have found limits to materials hence we need to think out of the box and find new materials.

Re:Asymptotic

[justforaday]

Are you referring to some sort of paradigm shift or something?

Re:Asymptotic

[abigor]

Well, not unless he's able to leverage it, because it's impacting the story we have to tell.

Re:Asymptotic

[Zocalo]

Without a major breakthrough, which isn't something I'd bet on, I'll agree that we are very close to the limits of silicon based CPUs. Strained Silion and Silicon on Insulator are effective stop gaps, but multi-core and possibly switching to something like Gallium Arsenide are the most likely ways forward for greater processing power at the moment.

Hard drives however? Some of the areal densities that are working in R&D labs are significantly denser than what we have now and will allow for plenty of capacity growth if they can be mass produced cheaply enough. Sure, we're approaching a point where it's not going to be viable to go any further, but we're not going to arrive there for a while yet. There is also the option of making the platters sit closer together so you can fit more of them into a drive of course. If you really want or need >1TB on a single spindle then I think you'll need to wait just a few more years.

Re:Asymptotic

[CmdrGravy]

I think it's going to take a lot of imagineering to fully appreciate the tectonics of a potential paridigm shift.

Re:Asymptotic

[Wordsmith]

Mods don't find marketspeak funny, apparently.

Re:Asymptotic

[lucifuge31337]

Without a major breakthrough, which isn't something I'd bet on, I'll agree that we are very close to the limits of silicon based CPUs.

Remember when 9600 baud was close to the limit of copper? Then 33.6. Then they changed how the pair was used, and made 128K ISDN. Then they changed it again and we're getting 7-10 MB DSL....sometimes even faster depending.

I find it hard to say the we're close to the limits of any technology in the computer/telecom field. Someone always seems to find a new way around it.

Re:Asymptotic

[ConceptJunkie]

Worse, they think it's insightful.

Re:Asymptotic

[Strange Ranger]

Like sands through the hourglass so go the chips on our dies. Thus, in order to birth a silcon sea change we need to get down to the granular level with the design schema.

Re:Asymptotic

[Thud457]

" I see all the mods are from the marketing department today. Either that or on crack... again."

Those two conditions are not mutally exclusive. Actually, they appear to be strongly correlated.

Re:Asymptotic

[rizzo]

Thankfully we've got some proactive synergies and tremendous upside.

Re:Asymptotic

[netwiz]

I find it hard to say the we're close to the limits of any technology in the computer/telecom field. Someone always seems to find a new way around it.

perhaps not, but things are getting really dicey WRT silicon processes. The lates process shrink to 90nm really hurt, and required bunches of tricks to make it work. Specifically, thermal dissipation is a big problem, as when you shrink chips, they get hotter, and require more idle power to make them work. This increases the total thermal power you've got to dissipate, and you've reduced the surface area with which to do so.

Leakage power is another problem. Sure, that 3.6GHz Prescott you've got there has a max dissipation of 110watt at full tilt, but it still consumes something like 53w doing nothing! That's pretty bad, and there's absolutely no fix for that. Physics and chemistry say so, and it only gets worse the smaller the transistors become. So 65nm will be a real bitch...

Re:Asymptotic

[Zocalo]

Yes, there's certainly a possibility that there may be a breakthrough, I just don't see it happening for several reasons. First and foremost we have the laws of physics; you just can't make the traces on the silicon substrate much thinner and still know for sure what's going on. This is something that strained silicon has alleviated a little, but without further size reductions then more GHz equates to more heat.

My other reasons are a little more subjective, but are largely to do with the fact that both AMD and Intel are investing heavily in developing multi-core CPUs. In Intel's case this has involved the very public scrapping of a promised CPU and a drastic revamp of its roadmap. While breakthroughs in CPU design have come from academia and other companies, the vast majority have come from Intel and IBM. However, neither are investing the R&D in ramping clock speeds ever higher and are focussing on multi-core designs instead.

Hence my original statement: based on what we current know about silicon based CPU design, we are at (or very close to) the limits of what is possible. Further R&D or a breakthrough might push that a little or even a significant amount higher, but without the massive R&D efforts of IBM and Intel, the chances of this happening are slim. Also, if the market does start to shift toward multi-core designs which seems very likely, then the inclination of people to look into better wats of doing things in the old way is likely to be reduced further.

Re:Asymptotic

[Waffle Iron]

Remember when 9600 baud was close to the limit of copper?

That was never the limit of copper. It was the limit of voiceband phone lines, which have artificially constrained bandwidth. Since voiceband is now transmitted digitally at 64Kbs, that's the hard theoretical limit, and 56K analog modems are already asymptotically close to that.

If you hook different equipment to the phone wires without the self-imposed bandwidth filters, then it's easy to get higher bandwidth. Ethernet and its predecessors has been pushing megabits or more over twisted pair for decades.

Re:Asymptotic

[arivanov]

No,

The lack of breakthrough will be due to something entirely different.

So far we have been exploiting the fruits of fundamental material science, physics and chemistry research done in the 60-es (if not earlier), 70-es and to a small extent in the 80-es. There has been nothing fundamentally new done in the 90-es. A lot of nice engineering - yes. A lot of clever manufacturing techniques silicon of insulator being a prime example - yes. But nothing as far as the underlying science is concerned.

This is not just the semiconductor industry. The situation is the same across the board. The charitable foundations and the state which used to be the prime source of fundamental research funding now require a project plan and a date when the supposed product will deliver a result (thinly disguised words for profit). They also do not invest into projects longer then 3 years.

As a result noone looks at things that may bring a breakthrough and there shall be no breakthroughs until this situation changes

Re:Asymptotic

[Casca]

Don't forget the symbiosis present with VARs and results oriented customers.

Re:Asymptotic

[AJWM]

Mod that +1 insightful.

I might also throw in the possibility that, since the end of the Cold War, there has been very little incentive for governments, etc, to back fundamental research that might (a decade later) lead to radically new technologies. Governments like the status quo, they like the future to be predictable. Fundamental research (except perhaps in really esoteric areas like cosmology or areas with practical benefits for them like medicine) scares the willies out of the people in power -- it might upset their apple cart.

Re:Asymptotic

[halivar]

If we put our brains together synergistically, and I'm sure we can reach a solution.

Re:Asymptotic

[bdcrazy]

Results oriented customers.... vs what? the customers that don't really care and will buy anything? oh, i get it. n/m me.

Re:Asymptotic

[scovetta]

For God's sake, please stop the business-speak!

Re:Asymptotic

[AJWM]

Sure, you'd need the chain down the 15K rpm SCSI version,

Heh, reminds me of those reports of possible antigravity effects with spinning superconductor magnets. Do you suppose if you manage to write the right bit pattern to every sector on the drive you could get it to lift off?

Re:Asymptotic

[Tanktalus]

Just to expand a bit on this. Not much - I'm going to grossly oversimplify this. Each "baud" is merely a change in signal. However, it is an analog change, not a digital change. These signals do not need to be either "0" or "1". They can be "2", "3", "4", etc. (there is a limit here, too, I'm sure). 33.6k is merely 3.5 times 9.6k, so we have amplitudes of 0 through 3 (4 discrete values, one of every two signals has an extra parity bit). Using 6 amplitudes (0-5), we get 57.6k, or, minus the parity, 56k. But we're still transmitting at 9600 baud.

Of course, that only matters to geeks. To the rest of the world, baud is irrelevant. It's how fast the pr0n downloads that counts.

Re:Asymptotic

[Tassach]

I don't know if you could scale a 15K rpm drive to use bigger platters. You would have a lot more centripital force at the rim -- I'd be worried about the platters warping.

Also, the linear speed might be too high to read without interleaving (which pretty much negates the advantage of the higher speed)

Some quick calculations:
Assuming that a 3.5" drive has 2.75" platters, which would have a circumferance of 8.64", would have a speed of 129,590 in/min at 15,000 RPM, which equals 122.7 MPH.

If we assume the 5.25" drive has 4.5" platters, these would have a circumferance of 14.14", which translates to 212,057 in/min or 200.8 MPH.

Also, the 5.25" platters are 268% larger (15.9 in^2 vs 5.9 in^2). Considering that the larger platters will also probably have to be thicker to prevent warping, an estimate of the platters having 3 times as much mass isn't unreasonable. This means much more powerful spindle motors, along with more heat, noise, and vibration.

None of these are insurmountable problems, but I doubt you could solve them economically enough to bring the unit price down so that it's competitive with smaller drives. The platter circumferance in the 3.5" drive is 8.64", which at 15,000 RPM is 129590 in/min, which translates to 122.7 MPH.

Re:Asymptotic

[halivar]

Mods don't find marketspeak funny, apparently.

Probably because it has nothing to do with Communism, old people, Beowulf clusters or setting up bombs.

Re:Asymptotic

[JWhitlock]

I might also throw in the possibility that, since the end of the Cold War, there has been very little incentive for governments, etc, to back fundamental research that might (a decade later) lead to radically new technologies. Governments like the status quo, they like the future to be predictable. Fundamental research (except perhaps in really esoteric areas like cosmology or areas with practical benefits for them like medicine) scares the willies out of the people in power -- it might upset their apple cart.

The government pumped over a half billion a year into the Human Genome project, and spent $1.6 billion on nanotechnology last year. The government is still willing to spend money on basic research, but I doubt they are willing to create a whole new agency, such as NASA. They would rather have private companies do the work (even if federally funded), then create a new class of federal employees.

I also think you are assuming malice on the part of the government, when instead you should be assuming stupidity. And, since it is a democracy, you don't have to look far to find the root of that stupidity.

Re:Asymptotic

[Kurt Gray]

I'll add that to my list of action items that way we can pick the low hanging fruit by the time we close the books on Q4.

Re:Asymptotic

[pthisis]

Computers on the desks of normal people has only been a reality for fifteen or twenty years. In that time we've moved from "A Typewriter With Pre-Print Proofreading" to "A machine that can do realtime interactive full motion video".

...and then stopped. 5 years ago. If we buy the 20 years figure, the last 25% of the PC era has had no need for faster processer speeds.

And until someone somes up with another must-have reason (a "killer app"), the demand for higher speeds simply isn't there. Somewhere around 200-500 Mhz, machines simply got "fast enough"--I remember the bad old days before then, when everyone I knew got a new machine every couple of years (or even every year). And it actually helped you with your everyday word processing, music listening, web surfing, spreadsheets, etc. But the last thing I needed a faster CPU for was DVD playback, and that hasn't been a problem for years.

Seriously, I'm a full-time programmer who does real-time music visualization as a major hobby,I'm enough of a geek to have run Linux exclusively for (literally) over a decade now on my desktop, and even I don't see a reason to upgrade my machine's CPU. For the majority of the public, the ever-faster CPU craze has been replaced by other needs. Lower power consumption, wireless, better peripherals/displays, handheld/music devices, etc.

I went from a 4 mhz 8088-> 20 mhz 386 -> 66 mhz 486 -> 200 mhz PPro. And you know what? I don't remember how fast the machines I've had since then are--my current one is a 1.3 or 1.4 ghz P4, I honestly don't know--because that's when I stopped caring. It just doesn't matter any more.

I think 1997 was the last time I bought a machine where I gave much thought at all to CPU speed. I haven't bought a new desktop machine in 4 years, and I don't foresee getting one in the next couple--but I have gotten handhelds, mp3 devices, etc. Indeed, the only reason I bought the last one was that my old one was very noisy, so I built a silent PC.

Re:Asymptotic

[SpecBear]

Fine, we'll take this offline. Just make sure to touch base regarding the status of your action items by EOB. We can't afford to lose momentum.

Heat is the problem

[CPNABEND]

Multi-processing is the way to go. We need to do that to help heat dissipation...

Re:Heat is the problem

[WaZiX]

The CPU spends as much as 75% of its time idle because its waiting patiently for the memory to give it something to do. With Systems only delivering information at a max of 1 Ghz and processors going up to almost 4 times as fast... Studies also show that they could in term be able to squeeze 20 Ghz out of wires as long as 20 inches (and only by 2010 will we be able to achieve that), but that would only be sufficient for the 32 nanometer generation of microships (and we're quite ahead of that)... So i think the future resides in optical connections within the motherboard, allowing processors to finally... well... process ;-)

Re:Heat is the problem

[dsginter]

Multi-processing is the way to go. We need to do that to help heat dissipation...

So, you think that using multiple iterations of an inherently power-hungry technology will somehow solve the power problem? While, certainly, we could back off clock speeds with multi-processing and reduce heat considerably, but, people always want the cutting edge so the demand to "crank it up" would still be a profitable venture, thus pressuring the price of the lower-end stuff.

Look at page 8. Processors are approaching the heat density of a nuclear reactor. Silicon is dead. We'll need something else if we want more clock cycles (or perhaps a new computing paradigm... something "non-Von Neumann).

Re:Heat is the problem

[Ironsides]

It is not cheap to move memmory onto the CPU. Cache is VERY expensive. And they already did that by moving the cache onto the CPU (it used to be an external chip like RAM). Moving ram onto the cpu would be very expensive and would limit system ability to adapt. More likely would be to move the memmory controller onto the CPU so that the CPU bypasses the system bus altogether when accessing memmory.

Re:Heat is the problem

[akuma(x86)]

A few problems with your post.

1) 75% idle time is nonsense. Where did you get that number? With SPECfp on an Athlon or P4 it's more like 20-30% idle. Just look at how spec scores scale with frequency to figure out the memory-idle time.

2) Increasing switching speed with optical technology increases bandwidth but does nothing for latency since nothing travels faster than the speed of light and electrons flowing along a wire can acheive close to 80% of the speed of light already. To reduce latency, what we need are smarter architectures and programmers that can prefetch the data into lower latency caches ahead of time.

Re:Heat is the problem

[WaZiX]

1) 75% idle time is nonsense. Where did you get that number? With SPECfp on an Athlon or P4 it's more like 20-30% idle. Just look at how spec scores scale with frequency to figure out the memory-idle time.

In an Article of November 2004s Issue of Scientific American about about optics-based computers.

2) Increasing switching speed with optical technology increases bandwidth but does nothing for latency since nothing travels faster than the speed of light and electrons flowing along a wire can acheive close to 80% of the speed of light already. To reduce latency, what we need are smarter architectures and programmers that can prefetch the data into lower latency caches ahead of time.

Huh, where did i even mention the speed of of electrons along wires? Im simply stating that wires will never be able to deliver enough data for the processor to be able to function at a correct regime.

Re:Heat is the problem

[NoData]

Electrical chips run far below 1% of c.

Yeah, the flow of electrons in wire is extremely slow, but the work is really done by the electrical field generated, so that as one electron is pushed into the wire, it "pushes" the sea of electrons forward so that an electron at the other end of the wire is shifted forward. This "shift" occurs pretty close to c. I

Don't complain.

[inertia187]

People in Soviet Russia, however, appear to be afflicted with amusing juxtapositions of the aforementioned situation.

Well Moore's Law is not a law...

[zoobaby]

It was just an observed trend. The trend is breaking, as far as retail availability, and thus we are not seeing our 10GHz rigs. (I believe that Moore's law is still trending fine in the labs.)

Re:Well Moore's Law is not a law...

[stupidfoo]

Moore's "law" has nothing to do with Hz.

From webopedia
(môrz lâ) (n.) The observation made in 1965 by Gordon Moore, co-founder of Intel, that the number of transistors per square inch on integrated circuits had doubled every year since the integrated circuit was invented. Moore predicted that this trend would continue for the foreseeable future.

Re:Well Moore's Law is not a law...

[jj_johny]

No, Moore's law was about price performance not about absolute performance. If you look at the cost of a PC it has consistently gotten better performance while decreasing in price. Nearer to the beginning of the PC revolution it was all performance inprovement and very little price drop. Then in the early 90s it was kind of balanced. Then the 2000 to 2004 was all about the machines getting cheaper with performance nudging along.

But now even you cheapest PC covers most users needs. So the CPU designers will continue to inovate but they will find that people will be able to keep their PCs and other electronics longer. Fundementally, the CPU business will start loosing steam and slow down. When people don't need to get new machines, they won't. The precieved premium for the high end products is getting less and less.

Engineering within limits brings great results

[skrysakj]

I remember the old days, when programmers nudged every
single bit of speed and capability out of the machines they had.
When computer engineers, faced with limits, still made magic
happen.

I hope this ushers that habit back into the profession. We have a lot of great technology, right now, let's find a better way to use it and make it more ubiquitous.

Re:Engineering within limits brings great results

[AviLazar]

So I can run four instances of CS and p0wn3d everyone in my single player, multi-character clan.
Viva la VM-Ware

Re:Engineering within limits brings great results

[Anonymous Coward]

but without a higher number of mhz on my CPU how will my penis get larger?

Re:Engineering within limits brings great results

[nurb432]

I doubt it will go back that way, we are to the point that they can be sloppy and get a way with it.

The limits are high enough now to not care. Back in the old days the limits were low enough that it did make a difference...

Not only that but the skills that used to exist in the older days are dissapearing.. "dont need to know that stuff'..

Re:Engineering within limits brings great results

[menkhaura]

These days we have processors hundreds times faster than 5, 10 or 20 years ago, we have thousands times more memory than we had yore... But, do our apps feel faster?

I tend to believe that they don't value efficiency and graceful code at all.

Re:Engineering within limits brings great results

[arkanes]

Er... yes? Do you actually REMEMBER how you used a computer back then? On my Windows 95 machine with 8 megs of ram swapping was slow enough that effective multitasking was out. On my modern spiffy computer I keep use multiple instances of enormous memory hogs of applications (that do stuff that nobody would have even considered adding to an application in 1995, like real-time analysis of a 50 meg C++ code tree), dozens of browser windows, all without flinching. At the article says, it's capability increase rather than absolute performance increase - there is no need whatsoever for a word processor that's faster than you can type. Speaking of that, in 1990 it was really easy to type faster than the Macs in our computer lab could keep up. So yes, MY applications certainly feel faster than they did 5 or 10 years ago. If yours don't then you either a) don't actually do anything that exploits the new power of your computer and your processor, and therefore don't need it or b) are looking at the past through rose colored glasses and don't remember when it took a weekend to run a compile, rather than it happening incrementally in the background without you even noticing.

Re:Engineering within limits brings great results

[gardyloo]

Ah, yes.
It seems that we need to review
The Story of Mel.

I'll post it here from several places,
So that the good people of /.
(and the other people of /.)
Don't wipe out a single server (yeah, right!)

http://www.cs.utah.edu/~elb/folklore/mel.html
http://www.wizzy.com/andyr/Mel.html
http://www.science.uva.nl/~mes/jargon/t/thestoryof mel.html
http://www.outpost9.com/reference/jargon/jargon_49 .html

and, of course, many other places.

Re:Engineering within limits brings great results

[grumbel]

### The limits are high enough now to not care.

The throuble is that this is assumption is wrong. The computers would in theory be fast enough to not care about optimization all over the place, the throuble is that a lot of bad programming doesn't result in just linear decrease of speed. If I use linear lookup instead of a hash-table, speed will go down, quite a bit more down then the amount of speed of the CPU increases over time.

Simple example, Gedit, an extremly basic text-editor takes 4-5 seconds to load on a 1Ghz Athlon, MSDOS edit on the other side on 386er started in a fraction of a second. From a feature point of view both do basically the same. Gedit for sure has some more advanced rendering and GUI and isn't a text-mode application like MSDOS edit, however shouldn't it be possible with todays CPUs which are quite a bit faster then back then to have an application that has better rendering then text-mode, but still be at least as fast or faster then back then?

Please, captain...

[salvorHardin]

...I cannae change the laws'a'physics!

Re:Please, captain...

[cnettel]

Bah, just use a tachyon burst followed by a concentrated tetryon beam.

Re:Please, captain...

[91degrees]

Anyone ever felt that was a bit rich coming from a TV series where the whole premise was based on changing the laws of physics?

Least of your worries

[dunsurfin]

According to most predictions we were meant to be enjoying lives of leisure by this point - working a 5-hour week in the paperless office, and driving to work in our hovercars.

Hardware resources and software design

[SIGALRM]

Make a CPU ten times as fast, and software will usually find ten times as much to do (or, in some cases, will feel at liberty to do it ten times less efficiently)

I find that software designers often do not take resource limits seriously. Programming is tedious, hard work. The algorithms chosen *are* important, and in some cases you shouldn't simply reach into the API toolbox and use the third-party solutions. There is no substitute for knowing how to write your own sort routines, specialized linked lists, and binary trees.

Re:Hardware resources and software design

[Derkec]

Right. But you also need to know when to write your own optimized software and when by using the API toolbox you won't cause much slowdown and will be able to deliver faster and cheaper.

I would also observe that programmer can be a lot of fun.

Re:Hardware resources and software design

[hng_rval]

There is no substitute for knowing how to write your own sort routines, specialized linked lists, and binary trees.

What about knowing how to use the libraries that have these functions built in, such as the stl? You might not be 100% as efficient with the libraries, but you can be sure that those libraries are tested and optimized, and if you write these functions yourself, they might be buggy and will most likely be slower than the what comes with the compiler.

Re:Hardware resources and software design

[gUmbi]

There is no substitute for knowing how to write your own sort routines, specialized linked lists, and binary trees.

Hogwash! Write first, optimize later...or in the real world: write first, optimize if the customer complains. Even then, what are the chances that I can write a better sorting algorithm than one included in a standard library that was written by some who studied sorting algorithms? Close to zero.

Re:Hardware resources and software design

[SIGALRM]

Hogwash! Write first, optimize later

No, you cannot retrofit quality and performance into a software project.

what are the chances that I can write a better sorting algorithm than one included in a standard library that was written by some who studied sorting algorithms? Close to zero

Maybe so, but it can (and should) be done in specific cases. For example, I maintain a library of binary tree functions, and I do use them frequently. They are well tested and perform beautifully. However, a project I completed recently required a large amount of data to be traversed in a specific manner, so we designed and built our own BTA--specifically optimized for the task.

As you know, poorly designed code will bubble up through the code and bite you in the end... and your project will suffer from it.

Re:Hardware resources and software design

[corngrower]

Hogwash! Write first, optimize later...or in the real world: write first, optimize if the customer complains.
Supposing that you need that first sale of your system to a customer, and when they demo your software, they see it's so slow that they dismiss it and buy the competitor's product. You don't have a second chance. This actually happened with a company I know of. The company pretty much went tits up because the architect neglected performance.

Even then, what are the chances that I can write a better sorting algorithm than one included in a standard library that was written by some who studied sorting algorithms?
I don't necessarily need to write the sort algorithm, but I need to be concerned with the effect of using the various algorthms on my system and select the corrrect one accordingly.
Again, that company that failed went with using a standard library for some functionality in the product instead of rolling their own and this had disasterous results. After the customer complained about performance, they found that they'ld need to completely redesign a significant portion of the product to correct the problem. It wasn't a two or three day fix. The fix would have taken 1-2 months. Try eating that cost when you're a small company.

Re:Hardware resources and software design

[ObsessiveMathsFreak]

No, you cannot retrofit quality and performance into a software project.

Quality does not nessessarily mean optimised code. For many customers it is more important to get code that works, doesn't crash and gets there yesterday. And if it's slow they'll either wait for an update or their next 10GHz PC! :E

Efficient code is a part of quality, but how important it is depends greatly on the customer. For video games and email servers, very important, for kernels... well we'd all rather a secure kernel that doesn't crash over an dodgy uber effient one. At least.... I would anyway.

Like all things it depends on what you want and what the programmer can realistically deliver.

And where is my Jetson's car!

[Evil W1zard]

There will always be points where technology slows down because it invariably will have to go through some total redevelopment instead of just building upon current products (like what they will be doing with the space program.)

A Good Thing?

[rdc_uk]

To my mind it _might_ be a good thing if the rampant speed-advance slowed (a lot).

Consider:

We might get some return to efficient coding being the norm, instead of writing systems anyhow and throwing more/faster hardware at it until it runs acceptably (Microsoft; its you I'm looking at!)

Your (and your business') desktop machine might _not_ become obsolete in no more than 2 years, and mmight continue in useful service as something more sensible than a whole PC doing the job of a router...

Processor designers might spend more time (i know they already spend some) on innovating new ideas, rather than solving the problems with just ramping up clock speeds.

Cooling/Quietening technology might have a snowball's chance in hell of catching up with heat output?

(and the wild dreaming one)
Games writers might remember about gameplay, rather than better coloured lighting...

Re:A Good Thing?

[nine-times]

Processor designers might spend more time (i know they already spend some) on innovating new ideas, rather than solving the problems with just ramping up clock speeds....Games writers might remember about gameplay, rather than better coloured lighting...

These both relate to a trend in the market that I believe we're seeing. Consumers are finding that their "old" computers from 2 years ago are still doing their jobs. When I have a 2Ghz Dell that I use for web surfing, word-processing, and e-mail, there's no benefit to upgrading to the newest 3.4 Ghz Dell. Though there's a hefty speed bump in there, most users will never know the difference.

Therefore, developers/manufacturers are being forced to focus on things like usability and features. They're making their products smaller and more efficient, easier to use, and making them fit transparently into the user's life better. They're focusing on the whole "convergence" idea.

Instead of people spending money on RAM upgrades, the money is going to smaller/lighter/better digital cameras, iPods, and home theater technology. In short, instead of seeing the same box being rolled out every year with better stats, we're seeing new boxes coming out every year with pretty much the same stats, but better designed boxes-- boxes that are actually more useful than last year's model, and not just faster.

I, for one, hope the trend continues.

Re:A Good Thing?

[akuma(x86)]

Oh my...where to begin.

>> We might get some return to efficient coding being the norm, instead of writing systems anyhow and throwing more/faster hardware at it until it runs acceptably (Microsoft; its you I'm looking at!)

Efficient coding is only useful if there is a return on your investment for efficiency. Exponentially increasing hardware capability over time at the same cost point makes this tradeoff obvious. The article is saying the hardware capability will still increase, but the programmer will have to learn to use concurrency to exploit it. This implies a fundamental shift in the single-thread world that we have lived in for a long time now.

>> Your (and your business') desktop machine might _not_ become obsolete in no more than 2 years, and mmight continue in useful service as something more sensible than a whole PC doing the job of a router...

The only reason a computer becomes obsolete is that something better comes along to replace it. Why else would someone spend money to replace something that's just as good. There must be value there. So, you're saying you want to stop or slow down progress??? Nobody is holding a gun to your head to upgrade. Keep your old computers and the rest of the world will move on and pay for increased value.

>> Processor designers might spend more time (i know they already spend some) on innovating new ideas, rather than solving the problems with just ramping up clock speeds.

Processor designers spend ALL their time innovating on new ideas. How do you think each and every new chip comes out faster than the last one? If we don't make a better chip, investors won't give us money to build new ones. It's all we do all damn day - build a better mousetrap. Innovation is part of the job. And "just ramping up clock speed" is pretty damn difficult thank you very much.

Re:A Good Thing?

[Eric604]

We might get some return to efficient coding being the norm, instead of writing systems anyhow and throwing more/faster hardware at it until it runs acceptably (Microsoft; its you I'm looking at!)

Microsoft's problem is memory usage, every unnecessary byte is processed at least once and is wasted cpu time. Ofcourse, reducing memory usage with huffman crunching won't make anything faster but the relation is clear: inefficient memory usage is a result of codebloat, non-streamlined datastructures, too many protocols/technologies and so on.

dual cpu systems

[Lawrence_Bird]

since the mid 90s thats all I have built - they really do extend the time before you feel compelled to upgrade. Sure there are not that many apps that run threads on each CPU. But to me a large part of it is that I run many applications simultaneously. With 2 CPU's I rarely get any sluggish feel. And if one app is being especially hoggish I can set it to run on one cpu and flip another important app to the other cpu.

This time around I also sprung for a hardware raid card and set up a 10 array. That has helped quite a bit with system responsiveness.

I've also turned off as much eye candy as possible. After a couple days its really not missed and things are much snappier.

yeah it would be great if I could run out and get some 10GHz chips to fry a few eggs on, but I think my dual MP2200's still have a bit of life in them.

Re:dual cpu systems

[t_allardyce]

With multi-cores and such soon all CPUs will be dual/quad CPUs. what would be even better is a hardware interface (like AGP is to graphics) for CPU's instead of this 'sticking it in a socket in the middle of the motherboard' approach, then you could just keep adding more and more CPUs and even keep the old ones (converter boards would let you plug in even older cpus). It would be pretty impressive getting it to work with different memory/cpu/bus speeds and having a fool-proof multi-cpu management system cheaply and in the basic PC specs but damn it would be cool.

Re:dual cpu systems

[Anonymous Coward]

Dude, with a system like that you should be able to move past the Courier typeface.

In the backseat of my...

[UncleRage]

flying car.

Where else would it be?

I've always wondered

[harks]

Why the size restraints on processors? Could a processor be made twice as fast if it could be made twice the size? When we hit the limit on how small transistors can be made, could processors continue to increase in speed by making them larger? I see no need why computers need to keep a processor size to two inches square.

Re:I've always wondered

[mikeee]

No, making it bigger will make it slower. Current digital systems are mostly "clocked" (they don't have to be, but that gets much more complicated), which means that signals have to be able to get from one side of the system to the other within one clock cycle.

This is why your CPU runs at a faster speed than your L2 cache (which is bigger), which runs at a faster speed than your main memory (which is bigger), which runs at a faster speed than memory in the adjacent NUMA-node (which is bigger), which runs faster than the network (which is bigger),...

Note that I'm talking about latency/clock-rate here; you can get arbitrarily high bandwidth in a big system, but there are times when you have to have low latency and there's no substitute for smallness then; light just isn't that fast!

Re:I've always wondered

[ZorbaTHut]

The problem with that is light speed. Transmitting a lightspeed signal across one centimeter takes about 3.3*10^-11 seconds - which sounds like a lot, until you realize that a single CPU cycle now takes about 3.3*10^-10 seconds. And I don't even know if electricity travels at true lightspeed or at something below that.

Another problem, of course, is heat - if your 1cm^2 CPU outputs 100w of heat, a 10cm^2 CPU is going to dump 1000w of heat. That's a hell of a lot of heat.

A third problem is reliability. Yields are bad enough with the current core sizes, tripling the core sizes will drop yield even further.

And a fourth problem is what exactly to *do* with the extra space. :) Yes, you could just fill it with cache, but that still won't give you a computer twice as fast for every twice as much cache - MHz has nothing to do with how many transistors you can pile on a chip. (Of course, you could just put a second CPU on the same chip . . .)

Re:I've always wondered

[Tacky the Penguin]

The problem with that is light speed.

Light speed is a big issue, but so is stray capacitence and inductance. A capacitor tends to short out a high frequency signal, and it takes very little capacitence to look like a dead short to a 10 GHz signal. Similarly, the stray inductance of a straight piece of wire has a high reactance at 10 GHz. That's why they run the processor at high speed internally, but have to slow down the signal before sending it out to the real world. If they sent it out over an optical fiber, things would work much better.

And I don't even know if electricity travels at true lightspeed or at something below that.

Under ideal conditions, electric signals can travel at light speed. In real circuits, it is more like .5c to .7c due to capacitive effects -- very much (exactly, actually) the same way a dielectric (like glass or water) slows down light.

--Tacky the BSEE

Re:I've always wondered

[rrowv]

> ...which means that signals have to be able > to get from one side of the system > to the other within one clock cycle. In a pipelined CPU (which accounts for nearly all in use today), it will take many many cycles for it to move from one end of the die to the other as the instruction executes. You're right, the bigger the die, the harder it is to have tight clocks if you spread everything out. But its just never done that way...

Re:I've always wondered

[jsebrech]

In a pipelined CPU (which accounts for nearly all in use today), it will take many many cycles for it to move from one end of the die to the other as the instruction executes. You're right, the bigger the die, the harder it is to have tight clocks if you spread everything out. But its just never done that way...

Very true.

Still, the signal needs to be able to cross the distance of the stage in your pipeline during the clock cycle. Smaller stages still mean you can have faster clock rates, as the intel chips demonstrate. All the clock rate benefits have come by making stages smaller (whether by reducing their functionality, optimizing their design, or shrinking the process). It seems we've reached the limit of how small they can be made, with intel seeing not just diminishing but vanishing returns of reducing the stage size to be able to bump up the clockrate.

Leave Moore's law out of this, please

[Raul654]

Moore's law has nothing to do with processor frequency. It says that semi-conductor capacity doubles every 18 monthsm, not frequency. (With the corollary that there is no appreciable change in price). As we all know, semi-conductor capacity is roughly proportional to speed, so saying processor speeds double every 18 months is not quite wrong, just a little inaccurate. On the other hand, saying that we're not seeing 10 ghz processors, so Moore's law is broken is wrong.

Re:Leave Moore's law out of this, please

[vykor]

Theories are explanations about phenomena, supported by evidence and observations. Laws are merely descriptions of phenomena. It's not as if you can eventually promote theories to law. They are two different types of things.
Moore's Law is a description of semiconductor packing and describes the phenomena of it doubling in a given time period. A Moore's Theory would be if it attempted to explain WHY this occurs.

Re:Leave Moore's law out of this, please

[nospmiS remoH]

On the other hand, saying that we're not seeing 10 ghz processors, so Moore's law is broken is wrong.

Er, huh?

** Preemptive response to my post follows **

Re:Leave Moore's law out of this, please (Score:3, Funny)
by Future_Child_Poster (723) on Friday January 07, @11:15AM)
You're new here aren't you?

Adding Ghz is probably not the best solution

[99BottlesOfBeerInMyF]

Ramping up clock speeds is hitting some serious limitations as far as increasing the work done by a machine is concerned. There are lots of ways to get work done faster. They are just harder to market without some good, popular, and independent benchmarking standards. At some point engine manufacturers realized that increasing the cubic centimeters of displacement in an engine was not the best way to make it faster or more powerful. Now most car reviews include horsepower. Clock speed is analogous to CCs.

Get over it

[Mirk]

If, as the Dr. Dobbs article says, "the free lunch is over", then the only sensible thing to do is make do with what we have now. For goshssakes, people, the computers we have now are already insanely over-powered. How many more gigahertz do we need my life already?

Re:Get over it

[Wordsmith]

Computers won't be fast enough until they can do anything we'd want of them near instantly. If I have to wait for feedback, it's not fast enough.

My Athlon64 3200, which isn't top-of-the-line but it's pretty close, still takes quite a bit of time to convert a DVD to divx. It takes a few minutes (because IO needs to get faster) to copy large volumes of files. Photoshop filters on huge, detailed files can take a few minutes to run. Machines only slightly slower choke on playback of HDTV. I can't imagine how long it takes to encode.

When I can do all those things instantly, do accurate global weather predictions in realtime and have my true-to-life recreation of the voyager doctor realize his sentience, THEN computers will be fast enough. Until the next killer app comes, of course.

Clock speed isn't everything.

[bchernicoff]

The difference between Intel and AMD's cpu architecture yields similar performance but at very different clock speeds(AMD's 3200+ runs at 2.2GHz). Other aspects of PC performance continue to improve, so as long as the trend is towards greater overall system performance, clock speed matters less. And greater parallelism is a good way to achieve this.

Abstract it away...

[Lodragandraoidh]

What gives?

You, sir, are an idiot. :p

Seriously though, the article recommends building applications concurrently. Short-term this may be the case on a small scale (and really already is the case).

The fundamental paradigm shift that will occur will be when we build our operating systems to handle concurrency for us; the advent of 4GLs will help move this forward.

In this model, you would program normally, not worrying about concurrency at all. The OS would do all the dirty work of breaking up your application into pieces that can run concurrently for you. Are we there yet? No. Will we be there? Yes - particularly if you want to keep productivity at high levels. You will have to abstract concurrency from the day to day programmer for this to happen.

Re:Abstract it away...

[arkanes]

The OS would do all the dirty work of breaking up your application into pieces that can run concurrently for you.

In a word, no. At least not with current languages. There's a reason we don't do this already, after all. Provably correct concurrency is very hard to generate, and almost impossible with pure machine code - you either end up with deadlocks and race conditions or very poor performance because you serialize too much stuff. Or incorrect results because data is transparently copied instead of shared. Etc. There do exist languages designed to accomodate and encourage both implicit and explicit concurrency, like Erlang, and I think we'll see more of them in the future, but it's not going to happen by simply ignoring it.

need for speed?

[ekeup1]

For >95% of users, I see no need to have computers faster than 2Ghz. Maybe I'm getting old... oh, and music these kids listen to....

Re:need for speed?

[grumbel]

### For >95% of users, I see no need to have computers faster than 2Ghz.

As long as there are games and a large number of computer users who want to play them, there will be a need for faster CPUs. While on the graphic side the main work is already done by the GPU, the physics and AI are still done by the CPU. And oposed to the graphics, where games are already quite advanced, AI and physics tends still to be rather primitive in games and will for sure need a lot of additional CPU.

bring on the diamond wafers

[AviLazar]

When they get off the silicon and hop onto those nice diamond wafers (there is an article in wired), then we will see faster processing.

The main problem - our largest producer (Intel) said they would not stop utilizing silicon until they made more money from it...We know that the industry likes to stagger upgrades. Instead of giving us the latest and greatest - they give us everything in between in nice "slow" steps so we spend more money. Personally, I wouldn't mind seeing the jumps of 1ghz at a time. This year 2.0 ghz, next year 3.0, following year 4.0, etc...and then eventually increase it further so its 5ghz at a time, etc. et al.

Actually this is sort of like competition

[melted]

When there's no free ride, programmers will have to compete with each other on who can squeeze that last bit of performance out of existing hardware. So you can kinda sorta predict the revival of the performance-conscious programming.

Two birds, one stone

[Tackhead]

> According to most predictions we were meant to be enjoying lives of leisure by this point - working a 5-hour week in the paperless office, and driving to work in our hovercars.

Judging from these pictures of the Intel retail boxed heatsink for the Pentium 4 560J (3.6 GHz), by the time we get 10 GHz PCs, the hovercar problem will take care of itself.

where is it?

[Fr05t]

"Based on decades of growth in CPU speeds, Santa was supposed to drop off my 10 Ghz PC a few weeks back, but all I got was this lousy 2 Ghz dual processor box"

Santa was unable to deliver your 10Ghz system this year for the following reasons:

1) Santa's Flying Car has not arrived

2) Santa could not use his sleigh because it failed the new FCC saftey requirements for subobital ships (something about flaming reindeer poo falling from the sky).

3) The OS for the new 10Ghz computer is Duke Nukem Forever which isn't currently available - maybe next year or decade.

Yeah

[Aggrazel]

And for that matter, where's my Mr. Fusion, Hovercar conversion, Jaws 17 and perfected weather service? Aren't those supposed to be done by 2015?

Your 10ghz is waiting..

[nurb432]

Just click here.. and send me your CC number, name and billing address ill get it shipped right out to you.

Free shipping if you act in 24hours..

But wait.. theres more..

Re:We need a faster bus

[mirko]

Maybe the guy who promised him a 10GHz PC was counting in binary ?

Moore's Law isn't Speed Doubling, it's Transistors

[unfortunateson]

The fallacy here is that the clock speed has to keep doubling. Moore's law says that the number of transistors on a chip doubles each 18 month period, and we're still pretty close to that.

Intel has just caved on the speed doubling in particular, by knocking the clock speed off their product designations, mainly because the Pentium M chips were running significantly faster than the same-speed P4's. AMD's Athlons have been 'fudging' their numbers by having the product number match not their clock speed, but that of the roughly equivalent P4 chip.

Meanwhile, cache sizes are up, instruction pipes are up, hyperthreading has been here a while, multi-core chips are coming down the pike... we're still getting speed gains, just not in raw clocks.

At the same time, the Amiga philosphy of offloading to other processors is truth, with more transistors on the high-end graphics processors than there are on the CPUs!

I hate to say it, but what do you think you need 10GHz for anyway? Unless you've got a REALLY fat pipe, there's a limit on how much pr0n you can process ;^)

The high-end machines do make good foot-warmers in cold climes.

GaAs and Relational Calculus

[Baldrson]

First of all, when DARPA decided to directly back specific technologies such as Danny Hillis' "Connection Machine" while supercomputer sales were flagging, they corrupted the market-driven support for supercomputing innovation. As a result just when Seymour Cray had a viable production line for GaAs cpus there was virtually zero market demand for the technology. The lower capacitance as well as higher mobility of the electrons of his version of GaAs technology weren't the sole benefits -- it was also about a factor of 10 cheaper to capitalize the fabrication facilities.

Whenever the government "picks winners" rather than letting nature pick winners, the technologists and therefore technology loses.

(Now that Cray is dead, according to the supercomputing FAQ, "The CCC intellectual property was purchased for a mere $250 thousand by Dasu, LLC - a corporation set up and (AFAIK) wholly owned by Mr. Hub Finkelstein, a Texas oilman. He's owned this stuff for five years and hasn't done anything with it.")

Secondly, as I've discussed before both operating system and database programming are awaiting the development of relations, most likely via the predicate calculus, as a foundation for software. Both are essentially parallel processing foundations for software.

This feeds into quantum computing quite nicely as well, as relations are not just inherently parallel, but are parallel in such a way that they precisely model quantum software.

Concurrency ... again ...

[C A S S I E L]

Concurrency is the next major revolution in how we write software

...as we've been saying for, oh, at least the last 20 years, which is about the time I was writing up my Ph.D. thesis on concurrent languages and hardware.

As far as I can see (being slightly out of the language/computer design area these days), concurrent machines and languages aren't taking off for the same reasons they didn't take off in the 1980's:

• Implicitly concurrent languages (ones where the concurrency comes for free) are either next to useless (since they tend not to have state, and have problems with a stateful world containing things like, oh, I/O), or end up not being very concurrent at all once they're running;
• Explicitly concurrent languages (ones with concurrency constructs) are tricky to program with, and debug, if you're trying to exploit the concurrency; shared memory (tricky at the hardware level) gives you multithreading, otherwise you're into the process world with very little in terms of shared objects etc.
• Concurrent hardware tends to have wacky constraints in order to operate with any degree of efficiency (Inmos Transputer anyone?) and is, again, a pain to program;
• The fancy concurrent hardware is custom-built, and by the time the boffins have built a concurrent machine that runs reliably based around processors of speed X, delivering concurrency of degree Y, Moore's Law dictates that you can go to your local computer store and buy a $1000 PC with processor speed greater than X * Y.

There's more than a handful of generalisations there, but in short: Moore's Law means that nobody is going to buy a highly concurrent computer when consumer PC's are still getting faster, and the people who really need high parallelism (modellers and the like) have their own special-purpose toys to work with.

Instead of asking about where it is

[EM Adams]

Start researching how to build better computers and start a company. You have some options to explore -

1) DNA/Molecular computers
2) Atomic switches
http://www.physicsweb.org/articles/news/ 9/1/2/1
3) Betacomputation (Switches made from neutrons and protons that can be on/off by adding/removing electrons bound inside of the hadronic structure)
This makes for good power supply too http://www.betavoltaic.com/
4) Positron/electron photon exchange
(Yes Virginia, antimatter/matter changes the phase of absorbed photons)
5) Integrated silicon/optic chips
6) Black holes (See Sci Am Dec 2004)

Also for all of you aspiring scientists out there do yourself a favor and join the present by reading about nonlinear/nonunitary mechanics
http://www.i-b-r.org/ir00018.htm
You'l l save yourself a lot of time by not modeling nonlinear processes with linear equations and infinite corrections under 10^-13 m (ultraviolet divergences).

Apple CPUs catching up....? WTF?

[TibbonZero]

And to think, that apple's CPUs are nearly at the same 'number speed' in the mhz race now!

Who would'a ever thought to see that happen?

What I need 10 ghz for

[way2trivial]

better than realtime video transcoding maybe?

authoring a DVD in less than an 4 hours from the dv-avi source?

my own CGI production in my lifetime?

Concurrent Applications are not The Answer

[smug_lisp_weenie]

There is one law in computer programming that is even more certain than Moore's Law: Over time, the user is going to do less work for the computer and the computer is going to do more work for the user.

Remember back when users had to wait in line in front of a terminal to run their punchcards through the mainframe? Back then, human time was cheap and computer time expensive. Nowadays the user's time is paramount.

Multithreaded programming breaks this law: It is hard to do multithreaded programming- Humans just don't think that way very well. To do it in a way that an arbitrary program (i.e. not a ray tracer) can see consistent performance gains in a multi-CPU environment is almost PhD-level hard. Making single-threaded software is already a major undertaking and anyone thinking that, in general, they should start designing all their programs as fundamentally concurrent programs is going to fall behind their competition due to other factors (security, features, etc.).

Instead, the only way concurrent programming is going to play a major role for the majority of software, I believe, is at the compiler and OS levels: The OS and compiler designers are going to have to do their utmost to transform single-threaded software to perform optimally in a multi-CPU environment- These folks are going to have to take up the slack that the slow CPUs clockspeeds are causing in terms of limiting the speed of Software- Concurrent programming at the application-level is only going to play a minor role in this, in my opinion.

Re:Should always specify North or South.

[BitchKapoor]

This means your CPU could be spread out over larger areas with little to no performance hit.

That's not true at all. At a mere 2GHz, light can only travel 15cm (6in) through free space in one cycle -- hardly a long distance. Add in modulation and switching delays, and you really can't ignore the board-level latency even with optical interconnect. On the other hand, even on-chip communication takes multiple clock cycles these days, so maybe it wouldn't be that much worse..?

Longhorn Screwed?

[SVDave]

According to Microsoft, an average Longhorn system will need to have a 4-6GHz CPU. But if when Longhorn arrives, 4GHz CPUs are high-end parts and 6GHz CPUs don't exist, well...I don't predict good things for Microsoft. Longhorn in 2007, anyone? Or maybe 2008...

GaAs??? GaAs is material of the future...

[PaulBu]

... and will always be! ;-) I think I first read this qoute sometimes in late 80s/early 90s, and it is still true. You know why? Ever looked at power dissipation specks of even the simplest GaAs chips? You would not want to build a processor out of those, Cray tried with Cray 4 and failed... ;-(

superconductors is the way to go for highest speeds/most concentrated processing power, due to extremely small power dissipation and extremely high clock frequencies (60 GHz for logic is relatively easy right now), but the problem is that after someone invests $3B in a modern semiconductor fab they do NOT want to build a $30M top-of the line superconductor fab to compete with it. IBM would be a good candidate for this, but they got burned on superconductor computer project back in 80s and would not touch it with 10 foot pole now, though both logic and fab has changed dramatically since then.

Disclosure: on my day job I do design III-V chips, and I used to design superconductor chips up until recently, now trying to push that technology forward is more of a night job for me... ;-)

Paul B.

Re:GaAs??? GaAs is material of the future...

[loose electron]

GaAs has a big problem with yield loss in manufacturing.

As such, it is ok for small stuff (under 20 transistors)but is not going to fly for million transistor CPU's

Re:GaAs??? GaAs is material of the future...

[k98sven]

superconductors is the way to go for highest speeds/most concentrated processing power, due to extremely small power dissipation and extremely high clock frequencies (60 GHz for logic is relatively easy right now), but the problem is that after someone invests $3B in a modern semiconductor fab they do NOT want to build a $30M top-of the line superconductor fab to compete with it.

I'd think the more likely reasons would have to do, for starters, with consumers not wanting or being able to afford a computer that requires constant cooling with liquid nitrogen (or even worse, liquid helium) to work.

Re:GaAs??? GaAs is material of the future...

[FuzzyDaddy]

Disclosure: on my day job I do design III-V chips, and I used to design superconductor chips up until recently, now trying to push that technology forward is more of a night job for me... ;-)

I haven't been in the superconducter field for ten years now... what's the technology being used for the switches/logic gates?

As for GaAs, it's alive and well in the world of RF (analog) amplifiers going up to 100 GHz - I think the current technology uses a 6" wafer. (see, for example, WIN Semiconductor)

Re:GaAs??? GaAs is material of the future...

[ChrisMaple]

Vitesse had CMOS GaAs as small as 0.35u and had to abandon the technology when smaller geometry silicon caught up in speed with GaAs. The money wasn't there (in 2000) to make a smaller geometry fab. Also, my understanding is that at smaller geometries the advantage for GaAs is reduced. Indium phosphide is another possible technology. The big problem is that a huge heap of money will be needed to develop a high speed, high integration replacement for silicon, and there's no guarantee that it will ever pay off. For the forseeable future, consumer processors will remain silicon.

Re:GaAs??? GaAs is material of the future...

[drinkypoo]

Given that a liquid nitrogen cascade cooling system is beyond the reach of most consumers, so-called "high temperature superconductors" are basically out of the question. Until we actually have cost-effective room temperature superconductors, I kind of doubt we're going to see much of this. Unless you mean something different than "superconductor" when you say superconductor, I'm at a loss as to where you are going with this.

Re:GaAs??? GaAs is material of the future...

[PaulBu]

I haven't been in the superconducter field for ten years now... what's the technology being used for the switches/logic gates?

Hmm, I am wondering what kind of logic were you using 10 years ago! ;-) Good old latching stuff? No, it was 1994, SFQ and Nb triulayer was already out there in the field, actually I did come to this country to work on it some time in '92, I guess...

Yes, it is SFQ/RSFQ (Single Flux Quantum) logic, counting individual magnetic flux quanta, but no, it has nothing to do with now over-financed "quantum computing". ;-) We can put tens of thousands Josephson junctions per chip now, all connected with matched superconductor transmission line (i.e., no RC time constants, nor F*CV^2/2 power), though which picosecond-wide pulse fly just fine. If you are interested, I can tell A LOT more -- hey, I'm one of the people who are still interested in pursuing this technology...

As for GaAs, it's alive and well in the world of RF (analog) amplifiers going up to 100 GHz

And with InP you can go to 150 GHz and maybe higher amplifiers (though not broadband), but there is a huge difference between being able to amplify a signal and being able to do any kind of meaningful digital logic at fixed power consumption... Actually, time for me to get off /. and get back to those pesky transistors... ;-)

Paul B.

Re:GaAs??? GaAs is material of the future...

[PaulBu]

STI and Conductus were successful in marketing PASSIVE HTS components (analog filters for cellular basestation receivers) and their main accomplishment was, actually, making "normal" systems engineers not to be scared of having a cooler in the system (providing a reliable cooler was also important ;-) ). The brilliant marketing gimmick was that they actually packaged a traditional filter and a switch in parallel with their SC filter in the same box, so if the cooler would fail the system would fall back to the traditional normal design, with some loss of capacity, of course, but at least it would still function.

As to digital logic, it is REALLY hard to make reproducible Josephson jucntions (active elements in SCE circuits) in HTS. One can make 2-4 of them for SQUID sensors (and it is a bit market for HTS too), but for digital stuff you need thousands and millions of them. In certain way HTS vs. LTS is similar to GaAs vs. CMOS -- it is easy to make a really nice, but simple, analog front-end in one, but the other can handle much more processing.

Replacing metal wiring on transistor chips with superconductor wiring will not help that much, yes, part of the RC constant which takes care of wire resistance, will be gone, but you'd still dissipate F*CV^2/2 power to charge/discharge the line. To fully utilize SCE logic one needs to use SCE active elements (current-sensitive JJs, not voltage-sensitive transistors).

I forsee the day that a user will be able to use a superconducting set of electronics on the desk.

Me too! ;-) It definitely can be done IF some larger system is built and verified first, then technology becomes a commodity. Check out, for example, this presentation by my former advisor and one of the godfathers of the whole field, seach for PeT workstation... ;-)

Paul B.

P.S. There is another fundamental reason to chose LTS, rather than HTS superconductors. The beauty of SFQ logic is that it uses almost quantum-limited amount of energy per switch. When one starts increasing temperature, thermal noise becomes too high (yes, even at 77K) and the main advantage -- tiny energy dissipation, which allows for very dense packaging -- goes away.

Re:GaAs??? GaAs is material of the future...

[14erCleaner]

You would not want to build a processor out of those, Cray tried with Cray 4 and failed... ;-(

I worked at Cray Computer, so I know something about this... The Cray-3 (also GaAs) was working OK when the bankruptcy hit, and the Cray-4 would probably have worked.

The failure of Cray Computer was due to competition and missing market windows, not due to the choice of technology per se. (Admittedly the late deliveries were due to difficulties with getting the processors working, but much of that was due to aggressive circuit-board designs that led to problems with open contacts, and the difficulty of repairing them).

btw, the 10th anniversary of the CCC bankruptcy is coming up on March 24th. My, how the years fly by sometimes.

Re:Moore's Law isn't Speed Doubling, it's Transist

[stratjakt]

I hate to say it, but what do you think you need 10GHz for anyway? Unless you've got a REALLY fat pipe, there's a limit on how much pr0n you can process ;^)


Photorealistic (or at least much better than the current high-end) rendering in real-time, I have some database apps that do a whole lot of number crunching, I have plenty of large projects that take 20 minutes to compile on a 3.06 P4 - CPU speed is the bottleneck on all of these.

A 10 Ghz CPU would probably bring with it 2GHz+ BUS and RAM.

You can never have too fast a CPU or GPU, too much RAM or too much HDD space.

A multicore CPU is great, but no substitute for raw speed. It's like comparing a bullet train to a fleet of honda escorts. The cars can move the same group of 1000 people, but the train does it so much faster and more efficiently.

Here comes the hertz gang again :\

[Mordaximus]

Which processor outperforms which:

1a)486-25SX
1b)486-25DX

2a)PIII - 450
2b)G4 - 450

3a)G3 - 300
3b)Playstation 2 - 300

Moral of the story : there are far, far more important performance measurements than clock frequency. If you think otherwise, you might as well slap a VTEC sticker on your case.

P.S. As other's have pointed out, Moors law has nothing to do cpu frequency.

Intel is to blame for this absurdity

[bigtrouble77]

This was spewed from Intel in 2002:

"First, by switching to the Pentium 4 architecture, Intel can drastically boost the clock speed. The old server Xeon topped out at 1.4GHz. The new one debuts at 1.8GHz, 2GHz and 2.2GHz, and will eventually pass 10GHz, she said."
http://news.com.com/2100-1001-843879.html

I can't find the exact quote and article, but another Intel exec/rep stated that this goal would be achieved by 2006.

Well, it's 2005, the P4 has topped out at 3.6ghz and has been discontinued because Intel has determined that the P4 arcitecture is streached to the limit.

Bottom line is that we should be expecting a 10ghz processor soon because Intel brazenly stated that they would produce one. Whenever they do make these statements the AP drools over the story, stock prices jump and I'm sure investors get excited.

Instead, their next gen processor is a 2ghz Pentuim M dothan. Intel should be ashamed of themselves for lying to the public and should be investigated for inflating their stock value though fictional claims about their processor technology.

Thanks to AMD, no

[gosand]

Processor designers might spend more time (i know they already spend some) on innovating new ideas, rather than solving the problems with just ramping up clock speeds.

Dude, that is what Intel was doing until AMD came along and forced them to get into this "keeping up with the Joneses" routine.

I can't decide whether to put a smiley face on this or not. I was being sarcastic, but for all we know it might be partially true!

Embedded Systems

[crimethinker]

We're already back to the "old days" in the embedded systems field, if we ever left. When you have to squeeze every bit of life out of a battery, you make sure your code doesn't piss away processor cycles. If the system calls for an 8-bit processor and a certain network daemon, you make the daemon run on the 8-bit processor, even if it was originally written for a 32-bitter. (then you whack the salesweasel upside the head for promising the moon to the customer)

If you crave the challenge of making tight, efficient code, sometimes with very little under you but the bare chip itself, then embedded systems might be the place for you.

cue the grumpy old man voice: "Why back in my day, we didn't have 64-bit multi-core chips with gigabytes of memory to waste, no sir, we had to write in assembly code for 8-bit processors, and WE LOVED IT!"

-paul

Another Flawed Law.

[twitter]

Quoth the author:

?Andy giveth, and Bill taketh away.?

That's only half right, because you don't have to let Bill take away. KDE3 runs well on a 233MHz PII and 64MB of RAM, almost a whole order of magnitude less of hardware than it takes to make XP happy. The picture is more drastic when you consider the virus load most XP setups must endure. You need a 2GHz processor just to keep running while your computer serves out spam and kiddie porn.

The changes Dr. Dobbs so wants are already happening in free software. There's a reason things like Arts can play music, games and system noises all at the same time while software on M$ has trouble sharing sound devices. If Beowulf is not a 10 year head start on concurrency, I don't know what is.

Quoth SVDave:

I don't predict good things for Microsoft. Longhorn in 2007, anyone?

Perhaps old Billy should put his money into processor development instead of SCO and FUD.

Actually, 56k is the hard limit

[Sycraft-fu]

Analogue lines aren't like DS-0 lines, which have a seperate control channel, the control is "bit robbed" from the signal. They take out 8kbps for signaling, giving 56k effective for encoding. That's why with ISDN there is talk of B and D channels. For BRI ISDN you get 2 64k (DS-0) B (bearer) channels that actually carry the signal. There is then a 16k D (data) channel that carries the information on how to route the B channels.

That's also why IDSL is 144k. The total bandwidth of an ISDN line is 144k, but 16k is used for circut switching data. DSL is point-to-point, so that's unnecessary and the D channel's bandwidth can be used for signal.

So 56k is as good as it will ever get for single analogue modems. I suppose, in theory, this could be changed in the future, I suppose, but I find that rather unlikely given that any new technology is likely to be digital end to end.

Re:You obviously haven't studied chip design

[Chemisor]

> You obviously haven't studied chip design

Perhaps that's why I am able to come up with a novel idea? Because nobody told me it's impossible, it just might work. But, of course, I welcome constructive criticism.

> How do you deposit another fresh layer of
> uncorrupted substrate on top of a processed layer?

With this technology it is already possible to do exactly that. It just needs a bigger nozzle.

> Chemical vapor deposition? It's not as easy as it sounds.

Neither was putting the man on the moon. But we did it anyway. Sure there will be engineering challenges here, but I see no theoretical problems with using CVD for this.

> What about thermal expansion/contraction?

Thermal effects in the sphere are no different from the ones in a flat plate. Also, there have been recent advances in painting transistors on flexible substrates, which could help on the surface layers.

> thermal effects on timing

How will they be any different from the ones in a flat CPU? Besides, you need to remember that with the clock in the center, timing is going to be far easier to implement.

> IR drop of a sandwich layer of
> substrate-oxide-metal-oxide-substrate-oxide-metal?

Perhaps you could explain this problem to those of us who don't understand the reference?

> How do you analyze process defects on the lower layers?

Just as you analyze process defects on flat CPUs: by testing them. I don't think chip manufacturers actually look at each chip under the microscope to see if something went wrong.

> If you want to do 3D, just make alot of chips and stack them together.

I don't see how that helps with anything. If you have flat chips anyway, why not just spread them out?

Re:Lying???

[bigtrouble77]

There's a big difference between a reasonable prediction and saying ridiculous things to inflate your stock price. I don't think it was reasonable for Intel to say, in 2002, that we will have a 10ghz part in the near future. Perhaps saying, 'Our goal is to reach 10Ghz by 2006', is a little more reasonable. But Craig Barrett and Co. don't talk that way (neither did Jerry Sanders of AMD). These statements could be looked at as devices to drive up stock prices. Finally, Intel said that the PIV's would hit 10Ghz. You can rest assured that's never going to happen.

But then...

[sczimme]

For God's sake, please stop the business-speak!

But then how are we supposed to leverage our synergies going forward to create a win-win situation? You are generating negative ROI in this incumbent conversation, and have become a cromulent addition to the team. You will be capsized^W rightsized immediately.

That humor isn't so far off

[GoClick]

Actually they have proactivly developed new thingys ;)

I'd like to note that the average 3Ghz PC can do MORE than the eqivalant of a 10Ghz 5Mhz 8086. Don't forget that it's not just your CPU doing math now days, it's that fancy $400 super-computer rivaling video card you've got too.

Re:Hertz don't put you in no drivers seat

[fm6]

Another beef is why they still have parallel, RS-232 and PS2 ports on monther boards. USB has finaly reached its potential, lets do away with all the legacy crap, there's lots of adapters on the market for people wanting to use their old stuff.

I think designers had the same idea a couple of years ago: you saw a lot of "legacy-free" systems with an emphasis on cheapness and tiny form factor.

But they didn't catch on for one simple reason: motherboards are a commodity. The pressure on price is enormous, so the only way you can turn a profit making them is to make a lot of identical motherboards, so you don't spend a lot of money on multiple assembly lines, or on retooling the lines you have between runs. So your cheap motherboards are a one-size-fits-all design -- and that means legacy ports.

Re:Lying???

[megalomang]

I fully agree that bullish statements by both companies were all over the place, especially stepping up after AMD beat Intel to the 1GHz punch, which also coincided relatively well with the added desparation created by the dot-bomb. The result is that as you pointed out, unofficial technology roadmaps to 4 years down the line were being quoted, compounding the possibility for gross inaccuracies.

But, I can assure you (I am part of the industry) that back then, the technology roadmap outlook was drastically different than today. It was impossible back then to understand the massive leakage issues at those speeds in 65nm and beyond since at the time, the warning signs were not unusual (i.e. they were overcome many times before on larger geometries). And believe me, the entire industry was practically blindsided by this. I think Intel was hit hardest simply because they were among the first to get there and were therefore aggressive on its adoption.

To say they were lying, and hold them accountable some kind of liability due to their confidence in 65nm would stifle future growth of the entire technology industry.

To single them out among all others who did the same would be unfair.

To even try to assign a dollar amount to this would be absurd. The entire industry took a beating at the same time. How much of Intel's stock plunge can be attributed to the failure of 65nm and frequency scaling promises? Is Intel not free to achieve these performance gains through other means such as core parallelism, memory architecture, higher levels of integration, and i/o architecture? Does this mitigate these dollar amounts?

The only stupid question

[starglider29a]

The one unasked:

Could we do a multi-processor system that splits the tasks according to their horsepower needs? The OS splits the tasks the processor that can handle it, an no more. Multilevel, Trickle UP CPU power. Say, for sake of argument, set up arbitrary CPU usage levels

1. A Pentium nothin' (or equivalent)
   relegate the slow, stupid stuff to a lesser processor, like Notepad, some sniffers... look at your Task Manager processes (or equivalents) for stuff you could be running on a 80286 and shove them to the CPU slums. This level keeps the lights on, controls the heat. It's the oil and the water pump on your Ferrari.
2. A Pentium ][ (or equivalent)
   Runs the OS functions themselves, if they can. File transfer, TCP/IP, simple Multimedia like MP3 & CD, Virus, Spyware detection... This level replaces the Pentium ][ in your kid's room running Limewire. It's the Stereo on your Ferrari
3. A Pentium ]|[ (or equivalent)
   The GUI, the heavy multimedia, like video, CD burning, the bloated web rendering... this level makes the UI responsive. It's the suspension on the Ferrari
4. A Pentium IV (or equivalent)
   The Engine of your Ferrari! The monster throughput... The Doom VII, the Celestia: Andromeda, SETI:Alpha 6, Climate Prediction, and of course the rendering of the perfect sig-other on Maya. When yer not using this part, shut it down, despin the fan and listen to the sound of silence.

The offshoot of this is that there can be a Level 5, if one were to plug into a cluster. From there, you can plug into BIGGER clusters until you either reach Blue Gene/X or you ARE part of the cluster.

Why not?