What's The Maximum Theoretical Clock For CPU's?

[Dash Lambda]

I've always wondered this, though I know it has very little significance in practice.

 

A tiny bit of quick arithmetic says that a signal traveling at c could cross a 6700K's die about c/.01352≈2.2174(10¹⁰) times per second, meaning it could cycle at ~22.174Ghz.

The signal in a CPU die doesn't travel perfectly at c, nor does it simply take a straight path across, so it'll be lower than that. Problem is, I'm just not entirely sure how I'd account for those things.

Anyone know where to start?

[Gruenbaum]

The cooling is the limit and the chips capability to handle that much current. You could be cooling the power of the sun under there, but the little connections that are in the CPU melt from the voltage

[MrSha256]

I'm pretty sure the electrons in copper travel at around 2/3 of c.

[SLAYR]

As fast as you can make a transistor switch, probabbly in the THz, you probabbly won't see that happen on a consumer chip in the next 5 years though.

Just now, MrSha256 said:

I'm pretty sure the electrons in copper travel at around 2/3 of c.

With electron drift, electrons actually travel really really slow, like cm/s or single digit m/s, because they is other electrons in their way.

[porina]

Similar thread recently.

[SLAYR]

4 minutes ago, Gruenbaum said:

The cooling is the limit and the chips capability to handle that much current. You could be cooling the power of the sun under there, but the little connections that are in the CPU melt from the voltage

Materials won't melt from voltage, that is like saying you will die from just being on a mountain. Voltage is like potential energy it is relative to where you are at. 

 

Voltage differences cause Current, which on the other hand will cause materials to melt because everything has resistance and Power (Watts) = I (current) *R(resistance)^2.

[Enderman]

4 minutes ago, SLAYR said:

As fast as you can make a transistor switch, probabbly in the THz, you probabbly won't see that happen on a consumer chip in the next 5 years though.

With electron drift, electrons actually travel really really slow, like cm/s or single digit m/s, because they is other electrons in their way.

No, because the amount of time for a signal to travel from one side to the CPU to the other takes a certain amount of time, and if you have a clock cycle run before the other side of the CPU is finished processing then you get computation errors.

 

http://www.design-reuse.com/articles/21019/clock-mesh-benefits-analysis.html

[Gruenbaum]

Just now, SLAYR said:

Materials won't melt from voltage, that is like saying you will die from just being on a mountain. Voltage is like potential energy it is relative to where you are at. 

 

Voltage differences cause Current, which on the other hand will cause materials to melt because everything has resistance and Power (Watts) = I (current) *R(resistance)^2.

Thanks!

[Dash Lambda]

7 minutes ago, porina said:

Similar thread recently.

Oh my, I thought I was more original than that...

Thanks for pointing that out~

 

9 minutes ago, SLAYR said:

As fast as you can make a transistor switch, probably in the THz, you probabbly won't see that happen on a consumer chip in the next 5 years though.

 

12 minutes ago, Gruenbaum said:

The cooling is the limit and the chips capability to handle that much current. You could be cooling the power of the sun under there, but the little connections that are in the CPU melt from the voltage.

 

I probably should've made it clearer that I'm talking about the theoretical limit of a perfect system. Absolute efficiency, no impurities, etc.

And the actual limit is how fast the signal can traverse its path, which includes the rate at which a transistor can cycle, but is not only that.

[Gruenbaum]

It really is a balance of the size of the chip, delay of message, and thermals. You could make a chip huge but it would be so slow

[Mira Yurizaki]

Keep in mind that signals in copper don't actually travel the speed of light in a vacuum. I don't know the exact figure but a EE colleague of mine said it was something like 2/3 the speed of light in a vacuum. This phenomena is known as the velocity factor, if you want to look it up.

 

But theoretically speaking it's the point where the switching happens so fast, too little current trickles down the line to the other transistors. You can open the voltage up but if you exceed the breakdown voltage, the entire system collapses. So... perhaps it's better to say it's more complicated than you think.

[Dash Lambda]

1 minute ago, M.Yurizaki said:

Keep in mind that signals in copper don't actually travel the speed of light in a vacuum. I don't know the exact figure but a EE colleague of mine said it was something like 2/3 the speed of light in a vacuum. This phenomena is known as the velocity factor, if you want to look it up.

 

But theoretically speaking it's the point where the switching happens so fast, too little current trickles down the line to the other transistors. You can open the voltage up but if you exceed the breakdown voltage, the entire system collapses. So... perhaps it's better to say it's more complicated than you think.

I did say the signal doesn't travel at c...

 

Though I didn't think about breakdown voltage... But I'm not actually sure if it would be reasonable to include that in a perfect model. I guess it would be reasonable since I am talking specifically about microprocessors. I just want to eliminate most of the real-world factors because, well, we already know the maximum clock for most modern CPU's in practice XP

[Mira Yurizaki]

2 minutes ago, Dash Lambda said:

I did say the signal doesn't travel at c...

 

Though I didn't think about breakdown voltage... But I'm not actually sure if it would be reasonable to include that in a perfect model. I guess it would be reasonable since I am talking specifically about microprocessors. I just want to eliminate most of the real-world factors because, well, we already know the maximum clock for most modern CPU's in practice XP

If you're going to eliminate the "real world factors", the answer is basically infinity.

[pyrojoe34]

44 minutes ago, MrSha256 said:

I'm pretty sure the electrons in copper travel at around 2/3 of c.

There's no solid rule like that, it depends on the thickness of the "wire". For example 12-gauge copper wire is 95% the speed of light.

 

if you're looking at an individual electron though it's like 80cm/hr, way slower.

[Dash Lambda]

7 minutes ago, M.Yurizaki said:

If you're going to eliminate the "real world factors", the answer is basically infinity.

Actually, I showed that a reasonable (if not high) upper bound is ~22Ghz.

I'm realizing that my question isn't very well structured in the first place. I'm not sure what factors I want to include, though I know I don't want to consider temperature.

 

I think the factors I want to eliminate are the ones we can continually improve, like efficiency. I don't know how much we can improve death voltage, however...

 

I don't think I know enough electrical engineering to even turn this into a reasonable question XP

[Mira Yurizaki]

40 minutes ago, pyrojoe34 said:

There's no solid rule like that, it depends on the thickness of the "wire". For example 12-gauge copper wire is 95% the speed of light.

 

if you're looking at an individual electron though it's like 80cm/hr, way slower.

It was for PCB signal traces or something like that.

 

16 minutes ago, Dash Lambda said:

Actually, I showed that a reasonable (if not high) upper bound is ~22Ghz.

That's assuming a signal has to travel across the die. There's nothing that says the rest of the system can't operate faster while this signal travels.

[Dash Lambda]

1 hour ago, M.Yurizaki said:

That's assuming a signal has to travel across the die. There's nothing that says the rest of the system can't operate faster while this signal travels.

Fair point, one thing that I was considering is that a core doesn't span the entire die anyway.