The Pace of Processor Speed

[johnnymk]

If I remember correctly, approx. 4-5 years ago, the speed race practically came to an end regarding processors. I believe the fastest clock speed available was the Intel 3.8 GHz Pentium 4.

If the manufacturers could have solved the heat problem, what do you think the clock speed would be today? 10 GHz or more?

[Kevster]

Clock frequency is really something that should not be used to compare processors, unless you're talking about processors of the same type and design. Ultimately processors should be compared by the number of flops/second they can process, and that is the direction we are generally going now. The key drivers in processor design now are: silicon process being used (smaller being better, as you can pack more gates in and perform more processing on the die versus sending it out to a support chip), the processing pipeline design (needs to be as efficient as possible) and the number of processing cores you can implement with that process, and the amount of on-die memory that is available for the processor to efficiently use. I know I'm probably missing a couple things, but those are the key points.

Right now, barring another great technology breakthrough, I don't think we're going to be seeing new multi-core 3.5+Ghz x86 processors in the next 12 months. Right now with Intel's 65nm process it has been estimated that 3Ghz is the ceiling for it's thermal design. Maybe with a new process that will different. There are dual core chips out at 3.2Ghz, and there are single-core chips out there above 3.5Ghz, but heat remains a key issue as it is directly related to a processor's clock frequency.

[zippyjuan]

Things seem to have reached the limit of what a single processor can do with today's materials. They are trying to increase speed more by adding more processing units instead of making each one faster.

[johnnymk]

OK, I know of the current limitations. But what if they were solved either by dramatically increasing the size of the processor or if some company magically solved the heat issue?

Again, considering at how rapid the pace was going 4-5 years ago, would they have gotten to 10 or 12 GHz by now?

Anyone will have to admit that the performance of processors has not increased at the rate that they were proceding back then.

[Jeffbx]

Well, the CLOCK SPEED has not advanced, but the power certainly has. Like Kevster says, clock speed is not a good measure of performance anymore, unless you're comparing 2 procs in the same family. But take a core2 & compare it against a Pentium 4, and the core2 is worlds apart in terms of performance, even though the clock speed is lower.

But to answer your question - if heat were not an issue, I think we'd easily be approaching 10Ghz or more by now. I'm still eagerly awaiting the day when diamonds will replace silicon in chip fabrication - at that point I think we'll see another significant increase in power & speed.

[shocky123]

They (Intel) make a 3.73Ghz version of the dual-core Xeon, but I hear its a power consuming monster. As far as AMD's side, I've heard that their new opteron will come in around 2.9ghz.

Where and how do diamonds become superconductors? Have a link?

[Markel]

Where and how do diamonds become superconductors? Have a link?
I think it's more the opposite - nonconductive and high heat tolerance:
http://www.pcworld.com/article/id,108295-page,1/article.html

[shocky123]

Thanks for the link, though I'm not entirely sure what they were planning on doing, at some level though, all crystaline structures (diamonds, silicon?) must have a lower limit on their workable size, just a shot in the dark, but wouldnt diamond be a difficult structure/compound to shape and mold?

It was an interesting read, wish they would have followed up with some more recent developments, I'll read up a bit on it sometime..

Also in development are 3-D silicon processes which somehow help power consumption.

Another note, 10Ghz isnt that hard to attain, there are optical switching fabrics in networking that can run at that, and 5ghz is cake, there are a TON of high performance NIC's these days that run at 5 or 10ghz. General purpose CPU's at that speed, we'll have to wait for that. There are a lot of improvements as others have said that can be made in efficiencies for cpu's, simply increasing the number of ops/cycle can make things super powerful. [see the 8800gtx, or other massively parrallel chips]

[Kevster]

Thanks for the link, though I'm not entirely sure what they were planning on doing, at some level though, all crystaline structures (diamonds, silicon?) must have a lower limit on their workable size, just a shot in the dark, but wouldnt diamond be a difficult structure/compound to shape and mold?


They could do it as a base substrate or a thin-film, and then use vapor deposition or ion-gun to dope it for the required properties it would need. The less conductive the better, especially for transistor isolation. That will allow you to pack more transistors into the same space and not have to worry about creep effects (flux, charge, etc).

[johnnymk]

BTW, what is the fastest processor (overclocked or not, water, nitrogen cooled, etc.) ever developed?

[Airencracken]

ion-gun

http://www.pbreview.com/pics/1108333917.jpg

?

http://images.wikia.com/starwars/images/e/e1/Planet_Defender.jpg

?

[shocky123]

IBM has demonstrated a 500ghz communications chip concept, and ran it at 354ghz at room temperature about 6 months ago. Its not a general purpose processor though.

I think I've seen a 6ghz x86 processor before somewhere.

[Kevster]

http://images.wikia.com/starwars/images/e/e1/Planet_Defender.jpg

?

:heh:

Actually, I was referring to Ion Implantation, which uses an electromagnetic 'gun' to accelerate phosphorus, boron or other ions to specific energies and impact them onto a silicon substrate. They will penetrate the substrate to a specific depth, which is quite usefull if you're trying to make a transistor on silicon.

Here's an info linky: http://silver.neep.wisc.edu/psii/

And another info linky with TONS of silicon processing tutorials: http://www.casetechnology.com/links.html (some of it is a bit dated, but it is a great informational resource site)