IBM builds 500 Ghz transistor

[DarkWaters]

What do you guys think of this: http://www.cnn.com/2006/TECH/ptech/06/20/i...reut/index.html

Perhaps this will not be as monumental as when Bill Shockley helped replace the vacuum tube with the first solid state semiconductor switch, but the news still sounds like this might be huge!

Any interesting thoughts?

[Olex]

It is definitely a good thing. I have personal two reasons however.

1. Multi-cores or multi-CPU suck. (Thank you very much Sony. That's right, we'll see the price you [sony] will pay for forcing programmers into your "kewl" PS3 architecture. )

Programming for them is a pain, and many aspects just can't be done well. Plus they still have severe achitectural issues with solutions for shared memory access for 2+ CPU's. Not to mention the debugging part and predictability of results are very low. (You never know the order of the execution.)

That said, multi-processing is the future, but it is not fully solved yet.

2. Single CPU's have been is a dead water for what? 3-4 years? I want to see Moore's progression. Maybe this breakthrough will help out here. B/c we've been stuck on under 3Gz for a while now. It's time to move on. Man, I'd love to have 500Ghz in my rig. ( [imagines what kind of physics would be possible to do in software] PPU will be nothing!)

----

It also sounds like their approach could work.

Most improvements in chip speeds over the years have come from shrinking the size of transistors, but IBM's approach is to tweak the silicon on the atomic level, meaning that transistors can be designed from the ground up with very specific applications in mind.

This sounds like a logical solution, though of course devil is in details.

EDIT: spelling

Edited June 22, 2006 by Olex

[jwoolcutt]

I just took a cursory glance at the article, and I could be mistaken, but I think there might be a difference between the measurements of Ghz for a _transistor_ and for a computer processor.

A hertz (and gigahertz) is a measurement of how many "operations" or oscilations occur in the device in one second. I've always understood that the measure of the Hz of a processor involves how many operations that processor could perform in one second, while measuring the Hz of a transistor (just a component!) would involve how many "toggles" that it could perform as a switch in one second.

So I guess I'd be more interested in seeing a comparison between the speed of this new transistor versus current transistors instead of to the speed of whole processors made up of millions of transistors. I imagine that the entire processor does not run at the same speed as a single component.

[DavidOdden]

I imagine that the entire processor does not run at the same speed as a single component.

No, but suppose a clock cycle were half that of the transistor speed. Yipes, that would be blazingly fast.

[DavidV]

I am ignorant of electronics, but I know that the clock speed of a CPU is the frequency of an oscillator crystal. Modern CPU’s have a frequency of around 3Ghz, or three billion cycles/second. Comparisons between the clock cycles of different processors are not valid because they have different instruction cycles, or the useful work that happens during each cycle. Adding 2+2 over and over is much simpler than a dedicated DVD decoding chip, which is much simpler than a complex 20+ step general-use CPU like the Pentium 4. This is why AMD’s Athlon CPU’s ran slower Ghz than a Pentium 4, yet were just as fast – they were able to do more work in each cycle.

1. Multi-cores or multi-CPU suck. (Thank you very much Sony. That's right, we'll see the price you [sony] will pay for forcing programmers into your "kewl" PS3 architecture. )

What are you talking about? Xbox 360 was the first console with a triple-core PowerPC processor architecture. The PS3 has a single 3.2 GHz Cell processor and 7 "Synergistic Processing Elements."

[Olex]

What are you talking about? Xbox 360 was the first console with a triple-core PowerPC processor architecture. The PS3 has a single 3.2 GHz Cell processor and 7 "Synergistic Processing Elements."

I'm aware of that. Xbox 360 does have 3 cores, with each capable of running two threads in hardware. However, one core is used for OS stuff, so you only get to use 2 of them, thus only 4 threads.

1. Xbox360 core are really nice solutions, since they don't force you to change much. You simply to get to run on two cores, like some dual-core Pentium D solution for PC. Xbox360 is a nice transition from current hardware without big flaws or hoops that developers have to jump through.

2. PS3 is a different animal. (all according to latest announcements by Sony) Yes, it does have one core (looks to be able to have 2 threads as well, not sure if one is taken for OS), but those 7 elements are a pain. In short, they have very little memory (256K) and have a limit on the execution code size (looks to be like graphical shaders on GPU).

So, basically, Xbox is like 3 big football players, versus 7 little guys with one team manager of PS3.

Xbox360 has a far better approach for its products (video games). PS3, however, requires a serious redesign of architecture to fit its 7 SPE.

Thus, I used PS3 as an example of multi-processing gone bad.

[DavidV]

There is an explanation of what "500 Ghz" means here:

http://arstechnica.com/news.ars/post/20060622-7117.html

[Olex]

Ah, thanks! This makes more sense now.