What is the ultimate limit to overclocking/manufactured frequency?

[pipnina]

So until recently I just assumed that this limit was temperature, because I had never looked into it in great detail, and I thought that once you were pushing enough voltage into a chip the thermals would eventually just become too much for even LN2 to handle and the proc/gpu/ram would overheat.

But in a recent der8auer video, where he overclocks all the 5950x cores to 5.8GHz with LN2, he mentions that actually you get to a point where the CPU cannot be stable at a higher frequency, and you can't make the CPU colder because it loses conductivity (I think he said -150°c~ for the 5950x). But in that case, why doesn't adding more voltage help like it did up to that point?

Is it just something to do with the laws of physics at that point? I'd have guessed so but then we have seen processors go above 8GHz before now on a single core.

 

What holds it back when everything you can do is already pushed way beyond the limits of practicality?

[Guest]

4 minutes ago, pipnina said:

So until recently I just assumed that this limit was temperature, because I had never looked into it in great detail, and I thought that once you were pushing enough voltage into a chip the thermals would eventually just become too much for even LN2 to handle and the proc/gpu/ram would overheat.

But in a recent der8auer video, where he overclocks all the 5950x cores to 5.8GHz with LN2, he mentions that actually you get to a point where the CPU cannot be stable at a higher frequency, and you can't make the CPU colder because it loses conductivity (I think he said -150°c~ for the 5950x). But in that case, why doesn't adding more voltage help like it did up to that point?

Is it just something to do with the laws of physics at that point? I'd have guessed so but then we have seen processors go above 8GHz before now on a single core.

 

What holds it back when everything you can do is already pushed way beyond the limits of practicality?

The design of the processor.

 

So in very very short.....

 

Main reason.

Long pipelines allow for high frequency. But this negatively effects per core IPC. That's why P4 and FX processors where slow, but could reach 8ghz.

 

Other Reasons

TLB errata

Cold bugged board or Cpu

voltage scaling.

and probably a few other things in minor relation.

 

 

[Fasauceome]

Some would argue that it's still temperature, and that the sub-surface components are getting too hot, but this is actually a misconception.

 

One factor is that electricity is an expression of kinetic energy, and there is a physical limit to how much kinetic energy a system can manage. You're stuffing electrons down really tiny tubes (in a certain sense) really fast so it's bound to have a physical limitation.

 

Electron flow also creates electro magnetic fields. Increased flow of electrons creates a stronger field, and this field itself induces an electron flow. This is how inductors work, in fact. A coil of wire (or other conductive substance) has electricity flowing in a circular pattern, which generates a current in the direction of the center of the coil. On a microscopic scale, the inductive effects can't quite be neutralized perfectly by designs of the chip.

 

There are many more reasons that are very significant in the speed barrier of CPUs, however. Not all of which have to do with the CPU itself, as @ShrimpBrime mentioned the motherboard is a major factor in the CPU performance.

[eeeee1]

I usually like to think of a PC analogy as a human. Overclocking ("stimulants and steroids") can push you to things that would have been extremely hard to accomplish without. But even if you "Overclocked" your body can only go so far (Heat, muscles, etc).

 

I know this analogy is extremely imperfect and someone is probably will get mad that I even mad it but hopefully this kinda helps.

[pipnina]

2 minutes ago, Fasauceome said:

Electron flow also creates electro magnetic fields. Increased flow of electrons creates a stronger field, and this field itself induces an electron flow. This is how inductors work, in fact. A coil of wire (or other conductive substance) has electricity flowing in a circular pattern, which generates a current in the direction of the center of the coil. On a microscopic scale, the inductive effects can't quite be neutralized perfectly by designs of the chip.

Ah, that kinda makes sense! Especially since IIRC magnetic field strength increases with the current, and these overclocked processors can end up drawing 300A+!

Not something I had considered.

12 minutes ago, ShrimpBrime said:

The design of the processor.

 

So in very very short.....

 

Main reason.

Long pipelines allow for high frequency. But this negatively effects per core IPC. That's why P4 and FX processors where slow, but could reach 8ghz.

 

Other Reasons

TLB errata

Cold bugged board or Cpu

voltage scaling.

and probably a few other things in minor relation.

 

 

A *long* pipeline? That kinda sounds like it would have the opposite effect to me as a layman. But then I don't know what that is so.
Thanks for the replies guys.

[Guest]

1 minute ago, eeeee1 said:

I usually like to think of a PC analogy as a human. Overclocking ("stimulants and steroids") can push you to things that would have been extremely hard to accomplish without. But even if you "Overclocked" your body can only go so far (Heat, muscles, etc).

 

I know this analogy is extremely imperfect and someone is probably will get mad that I even mad it but hopefully this kinda helps.

My stomach is the PSU. I feed it over regularly XD. Heart is the water pump. Brain is AI controller for temp and processing. Nervious system is NB and cpu I/O (HT or IF). Blood obviously my coolant. Eyes, obviously high resolution real time gaming!

[Guest]

8 minutes ago, pipnina said:

Ah, that kinda makes sense! Especially since IIRC magnetic field strength increases with the current, and these overclocked processors can end up drawing 300A+!

Not something I had considered.

A *long* pipeline? That kinda sounds like it would have the opposite effect to me as a layman. But then I don't know what that is so.
Thanks for the replies guys.

Information goes through the pipeline. There are multiple stages. Informaiton passes through the pipeline at a slower rate because it's longer. With more stages, the work is simplified for processing. The processor can than operate at a higher frequency, bringing AMD's FX processors to first to hit 5ghz from the box. But it was slower than Phenom II which was a completely different design with much shorter pipelines and actually better IPC at about 1000mhz less frequency. 4ghz Phenom could process that of a 5ghz FX. 

 

So with less stages, the Cpu transistors, are working a little more harder. It does depend on the instruction set as well. So with some overclocks that do not use AVX, you can overclock higher and obtain a score. Like WPrime vs Cinebench. 

 

In a nut shell. There's plenty more to it beyond my knowledge without looking on the interwebs for links and good reads.

Edited November 15, 2020 by ShrimpBrime
corrected less harder to more harder.

[Jurrunio]

1 hour ago, pipnina said:

because it loses conductivity

I wonder if he meant "thermal conductivity doesnt allow the cold to reach the transistors inside faster". Could be a German to English translate mishap