Super User is a question and answer site for computer enthusiasts and power users. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

I just wrote a script to throttle a heavy number-crunching program, to prevent the CPU from overheating. I can set any upper temperature limit, and successfully keep the CPU always below that limit. (Actually, my script predicts the temperature a few seconds ahead from current and recent temperatures, so it can throttle the main program before reaching the threshold, but this is a technical detail.)

Question: What should that limit be? What considerations should go into picking the limit?

Routine use of the main program always puts the CPU temperature in the 60-70C range. The "sensors" command from the lm-sensors package shows "crit" temperature of 95C. I assume this is the threshold which, when reached, causes an instant total power-down. Obviously, I want to stay far from that. How far? What I don't know if sustained temperatures of 80C is okay, or 88C, or should I keep it under 70C, or what. The lower the throttle limit, the less efficient my workflow, so I'd like to set it as high as possible without shortening the CPU's life or risking a power-down.

Answers to temperature-related questions on SuperUser, S.O. and various S.E. sites, and (dare i say) websites outside the S.O./S.E. world give some sense of normal ranges, but apparently for a CPU with a more leisurely lifestyle. My program can't do much if I keep the CPU temperature below 60C. Most people aren't theoretical physicists, it seems (shakes head, dumbfounded look on face). My heavy number-crunching program runs typically for one to fifteen hours.

BTW, this is a nice big desktop machine with plenty of good airflow. It never crashes, except after 10 to 20 minutes of solid number-crunching.

ADDITIONAL INFO: the motherboard is an Asus M4A78LT-M.

/proc/cpuinfo gives a ton of info times four (four cores) but the vital info seems to be:

model name  : AMD Athlon(tm) II X4 645 Processor
cpu MHz     : 3100.000
cache size  : 512 KB
share|improve this question
Can you please tell us the exact type of CPU? The upper limit heavily depends on the exact CPU type because laptop computers have other limits than desktop computers (e.g. my current laptop has a critical limit of 104°C). – Uwe Plonus Jul 8 '13 at 8:38
Three things: 1) You should not need to manually throttle your CPU. The CPU should do this completely on its own. The only reason to do this yourself is if you have a really old (P2/K6 era, ~ 1998) CPU without this feature or if you want to slow down the CPU before it gets hot (e.g. to reduce fan noise). There should be no need to interfere to keep things stable. 2) You can use to discover T-junction. 3) Do the crashes leave any logs (e.g. something else in the desktop can get to hot and fail. Hopefully leaving errors in a log). – Hennes Jul 8 '13 at 11:54
No logs are written. When it crashes, it's instant and total power-off. The CPU definitely does not throttle itself. Is this built into the chip and surrounding circuitry, or it done by a device driver, operating system kernel, or maybe in BIOS? This computer was built in 2011. – DarenW Jul 8 '13 at 16:48
If the system just shuts down without so much as a by-your-leave, then the cause is the CPU's internal thermocouple tripping its overheat safety, which in turn means that your system lacks adequate cooling. Solving that problem will give you a better result than a process throttling hack; with adequate cooling, you'll be able to run your processor flat-out for as long as you like without the safety tripping and shutting the system down. – Aaron Miller Jul 8 '13 at 17:31

You must log in to answer this question.

Browse other questions tagged .