What exactly makes CPU's GPU's clock further on Ln2/DICE?

[Lays]

Does anyone know the science behind why we can achieve higher clocks at the same voltages while on sub ambient solutions?

IE: CPU did 5 ghz 1.4v on Air, but can do 5.4 or 5.5 ghz @ 1.4v on Ln2.

 

Is it because the resistance of the silicon is lowered by the extreme cold?

 

If someone could explain it scientifically in layman's terms that'd be wonderful!

[Random]

Last i heard it had to do with caps superconducting at super cold temps. Their efficiency increases a ton

[K404]

There's a LOT of factors....which start to overlap with each other.

 

A digital clock rate is (simple terms) split into three... the rising edge, the time it spends in its "on" state (stable state) and the falling edge. A "movement" has to be stable for a system-specific amount of time on either side of a pulse (called set-up time and hold time) and nothing happens instantly, everything takes time.... just not much of it. it might take....e.g. 1ns to go from off to on. Or, if the pulse is being measured as a function of time, the rising edge is not vertical, it is very steep.

 

A "1" begins at e.g. 0.7 of the stable value (the threshold value.) The stable value contains excess, to make sure the signal cannot be misinterpreted. So, if a pulse is going from 0 to 5V (off to on) then the signal can be considered on from 0.7 of 5v, or when the signal gets to 3.5v The rest is "guarantee."

 

 

 

One effect of cold is that the thermal noise is reduced. That interferes with the quality of the clock pulse and how the "0" and "1" is interpreted. Reducing temperatures literally cleans up the signal. When the signal has "space to breathe" there is no problem, but as the pulses get closer together (frequency increasing) the "natural" noise that each one produces starts to overlap.

 

Increasing voltage (amongst other things) increase how sharply the rising edge of a clock pulse rises, so it takes less time to get to the minimum amount necessary to be seen as "on"

 

 

 

I bet someone just links to a webguide that is better written and more complete

Edited June 18, 2015 by K404

[Taloken]

In this guide from Sin about Z77 OC, he explain well the action of cold on overclockability :

 

http://www.overclock.net/t/1247413/ivy-bridge-overclocking-guide-with-ln2-guide-at-the-end

 

Take a look at the paragraph : The Science behind the 22nm 3D Transistor and how it can help us overclock! ^^

[ObscureParadox]

Also superconducting in CPUs is a no no since we need them to be semi conductors in order to function properly. Itherwise we will get a bunch of 0s or 1s coming through and nothing else.

[buildzoid]

You can drive higher voltages without risking the death of the transistors because the amount of loose charge carriers in a transistor decreases with cold. So at -200C a 32nm GloFo transistor will have a resistance of say 100K ohm when turned ON, at 50C the transistor will have only 70K ohm resistance when turned ON. This means that if you cool a CPU down the you can push much more voltage at lower or the same currents. This decrease in loose charge carriers with lowering temperatures is what causes the cold bug. If you cool a transistor too much there will not be enough loose charge carries left in the transistor to allow it to conduct when it turns on so even when it's ON it's effectively OFF.

Electrical current is what kills the copper wiring and transistors in your CPU. The other thing is that the copper wiring gets less resistive with cold which allows the electrical signal in the CPU to propagate through the wiring faster.

 

The above is what I know for sure from research into the subject. The following on the other hand is speculation:

I think that cold also lowers the gate capacitance of the transistors in the CPU this would allow them to change states faster because they take less time to charge up.

Edited June 26, 2015 by buildzoid

[dx4picco]

right, silicon resistance increases when temperature is lowered.

then "leak" are lower and stability is greater then you can push freq