5
\$\begingroup\$

Why does a processor overclock faster when cooled with liquid nitrogen? Also, is the inverse the reason why the processor slows down when it get's hot?

I'm specifically interested in Intel processors.

\$\endgroup\$

5 Answers 5

14
\$\begingroup\$

It's not the temperature that makes it run faster or slower. The lower temperature allows more heat power to be removed from the device, which allows more electrical power to be put in without burning it up, which allows it to be clocked faster. Part of the electrical power required is proportional to the clock speed.

\$\endgroup\$
16
  • 4
    \$\begingroup\$ @kurtnelle, many newer processors will automatically clock down if they get too hot to prevent damage. Most processors will also shutdown if they exceed some temperature (this could be anywhere from 70C to 95+ C) \$\endgroup\$
    – pfyon
    Jun 14, 2011 at 19:17
  • 3
    \$\begingroup\$ Because by slowing down the core can be run slower, which uses less electrical power, which produces less heat, which can still keep the processor from frying when it's hot. Actually I didn't know Windows was smart enough to do that. I'm taking your word on that and explaining the reasoning assuming it really does that. \$\endgroup\$ Jun 14, 2011 at 19:19
  • 6
    \$\begingroup\$ Now that I'm thinking about this more, it's more likely the processor hardware automatically scales back the clock when it gets hot than Windows having anything to do with it. \$\endgroup\$ Jun 14, 2011 at 19:21
  • 2
    \$\begingroup\$ @kurtnelle, David is talking about a minor effect. Regardless of whether the logic gates get a little faster or not a lower temperature, temperature directly does not effect the speed of a PC processor. That is governed exclusively by the clock speed. In other words, the processor runs as fast as we decide to clock it. Clock it too fast however, and it can burn up or won't work right. Cold allows it to be clocked faster, but by itself doesn't actually make it run faster. \$\endgroup\$ Jun 14, 2011 at 21:08
  • 2
    \$\begingroup\$ @kurtnelle Some cpu families have this function, core i7 series for instance has fixed TDP limits. Additionally the cpu is also much more susceptible to internal errors when running 'hot' or overclocked without extra voltage. Some errors are fatal (usually a BSOD in windows) some errors can be corrected but often require flushing the pipeline and/or caches which effectively halts the CPU for a period of time. Other errors aren't fatal, aren't caught by internal mechanisms, but are caught by the operating system which will usually attempt to correct for them, which can be a slow process. \$\endgroup\$
    – Mark
    Jun 14, 2011 at 21:49
5
\$\begingroup\$

You want to keep the core within its operational range. If you want to run the core faster which means it consumes more heat/energy then you need to remove more heat to keep it within its operational temperature range. If you can improve the cooling through various methods, fans, liquid, gas, otherwise, then you can add more energy in the form of an increased clock (multiplier). You can damage the part by trying to run it too cold as well as you can by running it too hot. Also there is a physical limit to this you cannot run it infinitely fast infinitely cold. Your limit is likely determined by the silicon itself, there is a limit to the clock multiplier, and you would have to change the reference oscillator to keep increasing the clock rate.

Think about the human body, try running a mile when it is 60 degrees outside, then find a place to try running in the same clothes when it is 110 or 120 degrees outside. Which one are you likely to make it through the mile without passing out (failing)? Which temperature will allow you to push yourself harder than normal? Keep your body within the operating temperature range and you can push it harder, at least for a period of time. If you warmed up inside before the run, you might push yourself a little faster if it were 50 degrees outside. But there is a limit, warm up all you want indoors, but in a t-shirt and shorts, you may not make it when it is 30 below outside, 40 below, 50 below.

\$\endgroup\$
4
\$\begingroup\$

Several folks have correctly answered that more cooling allows the CPU to run faster, BECAUSE if it is cooled it allows you to overclock it more. These same folks correctly implied what I am about to say plainly: This was not (until recently) a native talent of CPUs. And it never has been a physical law of semiconductors.

A CPU clocked at 1.2 GHz runs at exactly 1.2 GHZ, whether it is cooled to 60F or running at 160F. If you lose cooling (e.g., remove the cooling fan + heat sink), it will run at 1.2GHz until it melts itself into a puddle and can no longer run at any speed. But it will run at exactly 1.2GHz until the very second it dies. I'll bet many on this forum have actually witnessed/experienced this.

Some of the new-fangled computers have their own temperature monitoring and control systems which automatically enable/disable overclocking (or otherwise adjust CPU speed) based on CPU temperature. So if the CPU gets too hot, it slows itself down (reducing energy in) rather than burning itself up. (I reckon it does this by selecting a slower clock or by dividing the existing clock down; but I am not an expert on newer CPU innards.)

If the CPU cools down, the automatic governer circuitry reverses this process to allow the CPU to run faster.

\$\endgroup\$
1
  • \$\begingroup\$ At higher temperatures, both conductors and insulators are often less effective than they would be when cold. Consequently, many devices will start to malfunction when they get too hot, even before they have reached a temperature which would cause damage. \$\endgroup\$
    – supercat
    Feb 27, 2012 at 16:42
4
\$\begingroup\$

Just to clarify: It's not about just heat dissipation. You can dissipate 1000 W of power using boring water cooling, but it would not allow you to get to the top.

The idea is that property of semiconductors change, as well as resistance of interconnect (=copper). Lower resistance - lower RC constant, which is main factor limiting processor speed.

If one could cool it down to superconductivity stage, clock would increase even more - but this is unlikely for copper interconnect we see in current CPUs.

\$\endgroup\$
5
  • \$\begingroup\$ Also if you cool it to superconductivity, none of the transistors will work; there is a reason why it's called "semiconductor". \$\endgroup\$
    – user3045
    Jun 15, 2011 at 16:48
  • 1
    \$\begingroup\$ Silicon is not a superconductor, even at 0.000001K. Also, no luck with Copper - only Aluminum (which is another popular interconnect metal) is superconductor at 1.2K. \$\endgroup\$ Jun 15, 2011 at 16:55
  • \$\begingroup\$ So we should use Helium slush if we want the CPU to be really fast. \$\endgroup\$
    – Oskar Skog
    Apr 20, 2017 at 12:47
  • 2
    \$\begingroup\$ @OskarSkog Not necessary. Electron/hole mobility also gets lower at low temperature (it's just resistance is much more significant show stopper) - so there is an optimal temperature and it's not near 0K. Optimally you need something between 100-200K with specific temperature being different for different chips and manufacturing technologies. \$\endgroup\$ Apr 22, 2017 at 23:33
  • \$\begingroup\$ Ah yes, the negative temperature coefficient of semiconductors. \$\endgroup\$
    – Oskar Skog
    Apr 23, 2017 at 9:41
3
\$\begingroup\$

Semiconductors run faster at a lower temperature and at a higher power supply voltage. Higher voltage means more heat generated, which means that it has to be cooled more. The more you can cool a chip, the faster you can make it go.

\$\endgroup\$