Have computer speeds stalled?

They have yet to finish selling the 3GHz stocks at high prices.

All signs point to yes.

Are we not nearing the theoretical limit, anyway, owing to the capacitance between tracks and the fact that current starts jumping tracks at about the 5-6GHz barrier?

CPU progress has typically followed a stop-start pattern as new approaches are found to work (eg. someone figures out branch prediction, then everyone starts doing it and refining it, then progress stalls again as another issue becomes the limiting factor). We're well into a stall now, which means hopefully another breakthough any time now. :)

But yes, we're nearing the physical limits before random electron hopping starts to be an issue. Clock speeds aren't really going to increase any with the current system, so we've got to be smarter with them...

Quote:

Original post by SiCrane All signs point to yes.

Thanks for that link, that was a good article. I've just bumped threads to the top of my list of things to learn.

as about that barrier - i can say, for most of applications except games, rendering, science-related, etc., 3GHz is more than enough. In most of programs such as word processors, html renderers, GUI, etc., most of processor time is spend on instructions that is there only "because it is simpler to code that way".

For game/rendering/science apps, it is not a so big problem to do parallel computing. Take for example game physics. There is loop that iterates through all things in current state to compute next state. It can be done in several smaller loops running in parallel. Or rendering, set of triangles can be split into parts running in parallel. For raytracers concurency is not a problem at all, etc.

But try to imagine "word processor" or internet explorer running in parallel on several CPUs [grin] ... that's will be truly scaring, and i have no doubt that someday there really will be such things that will use some uberstrange programming techniques to concurrently do tasks such as parsing and rendering something like "HTML<++> 2020" [grin].

Quote:

Original post by Dmytry as about that barrier - i can say, for most of applications except games, rendering, science-related, etc., 3GHz is more than enough.

Hehe, I'm too lazy to find the quote, but your post reminded me of that quote where that guy 10 or 20 years ago said "I can't imagine why anybody would need more than [x low number] mHz of processing power." That's not an exact quote, but you get the point.

Quote:

Original post by Dmytry But try to imagine "word processor" or internet explorer running in parallel on several CPUs [grin] ... that's will be truly scaring, and i have no doubt that someday there really will be such things that will use some uberstrange programming techniques to concurrently do tasks such as parsing and rendering something like "HTML<++> 2020" [grin].

In task manager, internet explorer currently has 10 threads on my PC, MSWord has 3. They're already multithreaded.

Quote:

Original post by Dmytry as about that barrier - i can say, for most of applications except games, rendering, science-related, etc., 3GHz is more than enough.

It's not necessarily a bad thing, anyway. On the PS2/XBOX programmers have to be really good to get the games fast. But they get years on the same system to learn. If PC speed stops rising in such an easily exploitable way, programmers will ahve to get better at being efficient and knowing the architecture a little again...

Quote:

Original post by NewbJ Hehe, I'm too lazy to find the quote, but your post reminded me of that quote where that guy 10 or 20 years ago said "I can't imagine why anybody would need more than [x low number] mHz of processing power." That's not an exact quote, but you get the point.

I think Bill Gates said something like that, and it was about RAM ("Who would need more than 640K of memory?").

shmoove

A professor of me told during a class, that there's no technological barrier, we can go on with Moore's law for another decade, technologically speaking. However, the cost also increases exponentially, and it'll become economically impossible to produce much more transistors on a chip, unless all companies work together for the development and equipment, and specialize in what they use and sell these chips for instead.

Also, it took 8 years to develop the Pentium 4, with a team of 1000 people during the last few years. Try to imagine the cost of that. And newer processors with more transistors are more complex, hence require even more development.

The tools to be able to design chips (i.e. to decide exactly how the millions of transistors will be connected) don't evolve as fast as the ability to put more transistors on a chip.

Well, Intel cancelled their 4GHz cpu. Not because the architecture couldn't handle it, but because it became impossible to keep cool.

I wouldn't say clock speeds have stalled completely. AMD will most likely still work towards 3GHz, for example, and given time, it should be possible to reach higher speeds without generating as much heat.

But as has been said, there are plenty of other ways to improve performance.

yeah moores 'law', which ive always found retarded anyway, is most certainly losing credibility.

Quote:

Original post by benryves Are we not nearing the theoretical limit, anyway, owing to the capacitance between tracks and the fact that current starts jumping tracks at about the 5-6GHz barrier?

Apart from the heat, thats the second problem with our current technology. At some point (0.01 micron?), the electrons can jump from one circuit to another due to quantum incertainty, and the circuits don't mean anything anymore.

They will probably find ways to shrink the manufacturing process a bit further, which will make CPUs cooler and give them enough edge to raise clock speeds to perhaps, 8-10 GHz if we are optimistic, but its still going to end up stalling, and then it won't be possible to shrink the process anymore. Even there though, its a bit sad. Of course you can think "woah, so maybe if we get nitrogen cooled computers, we can have amazing 15 GHz processors". We possibly can, but 15 GHz is still not that amazing, especially if we can't get faster than that.

We're soon going to see CPUs with 2-4-8-16-32 cores that will hopefully increase performance.

I think what Dmytry's point should have been, is that having a 3GHz processor is already pretty useless if it's idling half the time waiting on the bus for I/O. I've noticed that the marginal improvement upgrading from 1 to 2 GHz is much, much less than upgrading from 486 to Pentium. Anyone here who can honestly say that their CPU is the bottleneck probably needs something massively parallel anyway, but for most applications programmers are spending much of their optimization time trying to reduce random access to memory and disk, because that's what really kills you.

In other words, there are other components in the machine which can be improved to make the whole thing noticably faster, whereas increasing the clock speed will give you very slight improvements.

Tom

Quote:

Original post by d000hg Quote: Original post by Dmytry But try to imagine "word processor" or internet explorer running in parallel on several CPUs [grin] ... that's will be truly scaring, and i have no doubt that someday there really will be such things that will use some uberstrange programming techniques to concurrently do tasks such as parsing and rendering something like "HTML<++> 2020" [grin]. In task manager, internet explorer currently has 10 threads on my PC, MSWord has 3. They're already multithreaded.

Hech, i have CPU with hyperthreading. When MSWord or MSIE or something else does something "CPU intensive" (that is, have lag for no reason) one half of my CPU is in iddle, one is loaded.
That's because only one thread does something, and other threads is waiting.
Even P1 166MZ would be enough for HTML viewing, if it would be done efficiently.

Moore's law is all about transistors. Nothing to do with frequency. Frequency is very close to barrier now, transistors is not. So, probably we will have more and more transistors (that is, paralelism) instead of higher and higher frequency.

486-->pentium versus 1GHZ-->2GHZ, i also experienced that. I played with 486 mainboard and pentium mainboard some time ago. Pentium 1 does so much more in single clock. There were no such step since P1. That is, drawing single pixel in my VoxelWorld on p3 takes as many clocks as on P1. And on P4 , significantly more clocks. SSE doesn't give any great benefits. With FPU , 2 additions can be executed in parallel. Same with SSE, and same for other commands except low-precision ones.

I think that instead of developing really faster systems, Intel switched to making higher and higher frequency at cost of making something else worse. AMD CPUs do more things in same number of clocks. And probably it will be possible to somehow boost AMDs frequency without increasing clock count, but it will take longer time.

Quote:

Quote: Original post by Dmytry as about that barrier - i can say, for most of applications except games, rendering, science-related, etc., 3GHz is more than enough. It's not necessarily a bad thing, anyway. On the PS2/XBOX programmers have to be really good to get the games fast. But they get years on the same system to learn. If PC speed stops rising in such an easily exploitable way, programmers will ahve to get better at being efficient and knowing the architecture a little again...

yes, probably. But there will be better compilers than for PS2 and similar things.

If you look at Figure 1 in the article, there are actually fairly similar plateaus in the late seventies and late eighties, which are hidden by a mass of low-performance dots and the fit to a straight line. Naughty, naughty.

Does anyone have any links of reversible computing other than that one AI dude's Artilect war article thing?

He said it was supposed to make chips run with little to no heat, and I want to see some research on this kind of thing... it is, after all, a prerequisite to 3D chip-er.... cubes. crystals? maybe?

Crystals. That'd be cool. Yay stargate!

edit: SiCrane, thanks for that article.

[Edited by - C-Junkie on January 11, 2005 11:06:36 AM]

Quote:

Original post by Dmytry Even P1 166MZ would be enough for HTML viewing, if it would be done efficiently.

True. I use my old 166MHz MMX box to browse the internet, listen to music, play Quake [grin], check emails and do WP - Windows 2000, IE5, MS Office XP - and it's perfect.

Quote:

Original post by C-Junkie Does anyone have any links of reversible computing other than that one AI dude's Artilect war article thing? He said it was supposed to make chips run with little to no heat, and I want to see some research on this kind of thing... it is, after all, a prerequisite to 3D chip-er.... cubes. crystals? maybe? Crystals. That'd be cool. Yay stargate! edit: SiCrane, thanks for that article.

I don't have any links on hand, however I did quite a bit of research on this subject for a class in cryptography, more precisely related to the effect of quantum reversible gates on computing.

The basic idea is: whenever we use a modern logic gate (and, or, etc.) with the exclusion of not there is a loss of information. The xor-gate, for example, takes 2 inputs and produces 1 output. Now, it is clearly impossible to know the state of the two inputs from looking solely at the output. If the output is 0 we know only that the two inputs are the same, however they could both be 1 or both be 0, and if the output is 1 we know the inputs are different. The same is true for or gates as well as and gates (to a lesser degree because we know if an and gate's output is 1 that both its inputs are one, however if its output is 0 we cannot predict the inputs).

Now, the way modern computers deal with this loss of an electron is to dissipate the electron as heat. Thus we have a limit on the theoretical speed of computing due to the fact that we can only dissipate heat at a certain maximum rate and this will be a limiting factor on computing speed.

The solution is to come up with a gate which has no loss of information between the input and output states that also performs a useful operation on the inputs (a two-to-two pass-through gate accomplishes the first but not the second). If there is no lost electron we have no heat that needs to be dissipated and the chips will run cool without cooling sources, thus removing the heat limitation on computers.

One example of a gate with these properties is the Toffoli Gate, it is a three-to-three gate with the following properties:

Inputs- A, B, C
Outputs- D, E, F

A->D
C->F
B xor (A and C)->E.

This gate performs a three-bit transform on the three inputs and produces three outputs from which you can predict the inputs.

The main problem which faces us, then, is finding reversible gates that have useful transforms associated with them, because the Toffoli Gate would be rather unwieldy to work with using current logical arithmetic because to reduce the output to its basic boolean components will require the use of lossy gates.

That's a very basic overview of the idea of reversible gates.