Does processor temperature really matter?

[RickC]

Does anyone have any good references about processor temperature? I don't mean temperature that is right up near the maximum temperature.

For example, let's say a particular processor has a maximum specified temperature of 65C. Let's says it's running Prime95 all the time. Does it make any difference if the processor is running at 55C or 45C? I mean like is it going to last longer if it is cooler or make less mistakes if it is cooler?

I know this type of thing is not always obvious. With hard drives there are articles that say cooler hard drives have less failures and other articles that say warmer hard drives have less failures.

Thanks,

Rick

[xilman]

Originally Posted by RickC

Does anyone have any good references about processor temperature? I don't mean temperature that is right up near the maximum temperature.

For example, let's say a particular processor has a maximum specified temperature of 65C. Let's says it's running Prime95 all the time. Does it make any difference if the processor is running at 55C or 45C? I mean like is it going to last longer if it is cooler or make less mistakes if it is cooler?

I don't have references immediately to hand, but I will be astounded if running at higher temperatures doesn't shorten the expected life. Simple consideration of atomic mobility as a function of temperature leads me to that conclusion.

The question you should be asking, IMAO, is whether the lower temperatures make any practical difference. For instance, if the lifetime is reduced from fifty years to twenty years, do you actually care?

FWIW, I overclocked a PPro from 180MHz to 200MHz without changing the cooling in any way. The system ran 24/7 for about 15 years in a variety of environments without air conditioning. As far as I know, it still works perfectly. The only reason it's not running now is that I've other machines which are more cost-effective to run.

Summary: don't worry, be happy.

Paul

[ATH]

My experience is that it runs slightly faster the lower you can get cpu temperature, down to about 40ºC, lower than that don't give much speed I think.

When my old computer clogged up with dust and got up to around 55-60ºC the iteration time in Prime95 went up, and cleaning it out brought the temperature and iteration time down.

[lavalamp]

I have no sources, but I seem to remember that cooler processors live longer, and warmer processors make fewer mistakes.

I can't think why that second part would be true, so perhaps someone else could lend a more technical eye to that claim.

[mdettweiler]

Originally Posted by ATH

My experience is that it runs slightly faster the lower you can get cpu temperature, down to about 40ºC, lower than that don't give much speed I think.

When my old computer clogged up with dust and got up to around 55-60ºC the iteration time in Prime95 went up, and cleaning it out brought the temperature and iteration time down.

In my experience, speed is only affected over a certain temperature value (often referred to as Thermal Spec, and usually posted on the manufacturer's website). For intance, on my Core 2 Duo E4500, it begins thermal throttling at 72 C. (Note that your "observed" throttling threshold may vary due to inconsistencies between temperature-reading programs' readouts--for instance, Core Temp usually gives me different results than RealTemp.)

[henryzz]

I am pretty certain that when reading about overclocking someone couldn't get his machine Prime95 stable until he reduced the voltage so that the temperatures went down.

[Xyzzy]

We do not sweat temperatures too much. Our AMD Phenom quads run 63-65°C when they are dirty and 57-58°C when they are clean. They have a published maximum of 75°C and throttle at 73°C.

Even though the temperatures are a bit high, the boxes return valid work, including, recently, a factor for F₁₂.



We worry more about the effects of thermal cycling. Our quads are not connected to the Internet so we never bother with updates and we never have to reboot. The total power cycle count for each box is in the single digits, and we have had them since December 2008, running 24×7. They have battery backup protection.

Now, if someone ever decided to toss GMP-ECM 6.2.2 into Debian "Lenny" (stable) we'd update even if it meant killing our triple digit uptime.

http://packages.debian.org/search?keywords=gmp-ecm

[Colt45ws]

My i7 runs 65-71C depending on room temp.
This is a 920 being run as hard as I dare. Im almost at maximum recommended voltage to get 4.0GHz.
Something I noticed when I was fine tuning it, it needed a bit more voltage when it got hot. I shut the A/C off and let all the machines heat this place up. I let it go as far as I could, about 30C. Internal case temp was 45C and the proc was running close to 85C. The voltage I was at I had already vetted with multiple programs, but with the machine cooler, under 38C internal temp. With it that hot, it wasnt stable at all until I bumped the voltage a notch.
Seemed odd to me initially, but I guess resistance increases as it gets hotter and it needs more voltage. I can look at the wattage being used by the processor on this neat little utility. The extra temperature was making it use an additional 7 or 8 watts over the cooler temperature. Does not seem like much, but I guess if its _barely_ stable it might cause it to need additional voltage hot.

[lfm]

Originally Posted by RickC

Does anyone have any good references about processor temperature? I don't mean temperature that is right up near the maximum temperature.

For example, let's say a particular processor has a maximum specified temperature of 65C. Let's says it's running Prime95 all the time. Does it make any difference if the processor is running at 55C or 45C? I mean like is it going to last longer if it is cooler or make less mistakes if it is cooler?

I know this type of thing is not always obvious. With hard drives there are articles that say cooler hard drives have less failures and other articles that say warmer hard drives have less failures.

I believe the theory is cooler is better because heat makes the metal traces develop tiny filaments that eventually will cross another trace and cause the system to malfunction or die. The more heat the faster they grow but even moderate within spec temps will slowly grow these hairs. The result is random and can take a long time to cross another fatal place or a short time.

Sorry I don't have any scientific references, this is just from hearsay.

[joblack]

Originally Posted by lfm

I believe the theory is cooler is better because heat makes the metal traces develop tiny filaments that eventually will cross another trace and cause the system to malfunction or die. The more heat the faster they grow but even moderate within spec temps will slowly grow these hairs. The result is random and can take a long time to cross another fatal place or a short time.

Sorry I don't have any scientific references, this is just from hearsay.

In theory that's right, in pratice it takes 15 - 30 years to create problems. In that time you have upgraded to a newer version.

The problem with higher temperatures is that your power consumption will raise. That means you pay for more energy which is used less efficient.

It depends on what you want - most bang for the energy or maximum bang at all ;).

[lfm]

Originally Posted by joblack

In theory that's right, in pratice it takes 15 - 30 years to create problems. In that time you have upgraded to a newer version.

The problem with higher temperatures is that your power consumption will raise. That means you pay for more energy which is used less efficient.

It depends on what you want - most bang for the energy or maximum bang at all ;).

OK, I guess a more immediate problem would be any flaws in production where a trace is too thin or has some impurity could heat up and actually melt like a light bulb filament. The more voltage, current and heat you feed in the more sensitive to any poor connection it would be. And since there are a lot if thin little wires in a chip the chances of an poor trace are non-zero. Just like a light bulb, the problem could take a while to "blow out".