Does heat affect the speed of a CPU?

[AmigaMance]

I came across to this thead here: http://www.amiga.org/forums/showthread.php?t=1299                                    
In which a user claims that he got a big speed increase from his ppc CPU, by replacing the heatsink with a bigger one.. :-)                                                  
Is this claim, valid? I know that heat affects the stability of a CPU and has an extremely small impact in its speed, but this is the fist time that i read something like this.

[Framiga]

you can bet! its much more than valid.

This applies on some other components too (ram controller/ram sticks, logic, scsi chips etc).

[Ilwrath]

you can bet! its much more than valid.

The statement that heat alone can have a direct influence on speed?  Are you sure?

From what I've seen, I'd highly disagree.

Heat certainly affects stability and lifespan of the chip/device.  But I've not seen a direct correlation to speed, with the noteworthy exception of some modern processors which throttle speed back when a certain thermal threshold value has been exceeded.  As far as I know, the PPC chips used back then had no such technology, and therefore shouldn't gain or lose a significant amount of raw speed based upon temperature.

Now, of course, what might have been meant is that if the device was running too hot, it might have had some smaller glitches and problems occurring, which were adversely affecting the apparent performance.

[Framiga]

my direct experience about various electronic gears, says yes.

We are speaking about of not only CPUs but as already said, other tied components as well.

If you don't need a heater, cool down heat in electronics is one of the main problem and yes it slows down speed. (other than reduce the lifespan-stability as you said)

[koaftder]

AmigaMance wrote:
I came across to this thead here: http://www.amiga.org/forums/showthread.php?t=1299                                    
In which a user claims that he got a big speed increase from his ppc CPU, by replacing the heatsink with a bigger one.. :-)                                                  
Is this claim, valid? I know that heat affects the stability of a CPU and has an extremely small impact in its speed, but this is the fist time that i read something like this.

Temperature wont affect speed of a clocked system in a general sense. The ic's have an optimal temperature range. Being in the optimal range will give you the best proformance. Getting to hot will cause undefined behavior. A fault tolerant system can overcome problems, but you will suffer proformance.

A stock amiga 1000 will not run software faster when dipped in liquid nitrogen than when running at room temperature for example.

I'm not sure what dipping an a1000 in liquid nitrogen would do... Would it raise the propigation time accross the gates? How would it affect the resonators? Somebody needs to try this.

[Chunder]

Not sure about the PPC family, but some processors have thermal throttling, whereby if the processor gets too hot it doesn't just pack in entirely - it slows down.
Thus, increasing the cooling of the processor will return it to the full speed.

[AmigaMance]

 From my personal experience, i agree with Ilwrath and koaftder. I haven't notice any corelation between temperature and CPU speed. That doesn't mean that Framiga is wrong, maybe he has experienced something that we haven't.

@Framiga
 Are this changes measurable from a benchmarking utility, like SysSpeed for example?

@Chunder
 No, as Ilwrath said, i don't think that this is the case with PPCs.

[minator]

Making a CPU (or any chip) too hot will cause it to operate improperly or fail.
Cooling it will make it operate properly.

Cool it sufficiently and you may be able to raise the clock speed *that* improves performance.

[PaSha]

IIRC NMOS chips go faster the hotter they get, increasing the heat even more, thus going faster again and so on.... until they die.

I know one kind of chips do this, I think it was NMOS, but I'm not quite sure...

-Paul

[Daedalus]

Yup, NMOS chips did suffer from thermal runaway, but they're very very rare nowadays, and in any case, modern chips are generally throttled back by external clocks so they couldn't suffer this fate. Some NMOS chips used to use internal clock generators or frequency multipliers/dividers, and as they heated up, propagation times dropped, making them run faster which generated heat, and thus creating a spiral of doom for the chip ending in burning out the core. CMOS chips are the opposite, propagation times are longer with higher temperatures so this can't occur, but the fact that they're nearly always externally clocked anyway makes it a moot point.

[KThunder]

the speed of a cpu is controlled by a crystal, crystals are unaffected by temperature.
temperature limits how fast a given componant CAN run because of overheating.
a 3.0ghz athlon running at 100f is exactly the same speed as the same cpu at 150f
increase the speed however until the cpu overheats and shuts down and then yes the temp slows it down to 0ghz
this is another consumer confusion area. confusing overclocking and speed and temp etc.

yes some mobile cpus do throttle themselves due to power needs but afaik none do because of temp.

[Framiga]

@Framiga
Are this changes measurable from a benchmarking utility, like SysSpeed for example?

sorry AmigaMance but with all my respect, i have NO intention to leave out the CPU fan to perform benchmarks . . eh! :-)

[Stedy]

The hotter a semiconductor device gets (CMOS technolgies) the slower it goes. Take a modern CPU/FPGA to -40C and it speeds up it's timing by 15%. This is basic semiconductor physics and Ohms law.

This does not mean a 15% system increase, the timing from the CPU to the peripherals is 15% quicker. If the peripheral is also cooler, it could run quicker. I design systems to work from -40 to 85C ambient, we know from experience they are quicker at -40 than 25C.

Unless you cool your entire system to below 0 C you will not see a significant speed increase. Surprisingly a number of manufacturers keep the affect of temperature on speed under NDA!

HTH

Ian

[BenShep]

The intel p3 series has thermal throttling.