Temperature vs Execution Times

[pheonixashes11]

Hi,

I am trying to find the relation between temperature and execution times for processes like FFT, Sorting and Matrix multiplication.
For this I have started these processes each from different starting temperatures like 50, 55, 60, 65 and 70 (in centigrade).
It was observed that the execution time decreased when the temperature was high( when a process was started from starting temperature of 65 or 70).
What can be the explanation for faster execution times at higher temperatures?

Note : I am using Intel Core 2 Due E6420 processor. Frequency scaling is turned off while execution.


Thanks and Regards,
Bijju Kranthi V

[Rubberman]

The more load the cpu is undergoing, the hotter it will get - to a point. I monitor my system temperature, both CPU and RAM, and have found that under high load conditions, such as when building a new kernel with 8 make jobs ( make -j8 ) that pegs all 8 cores on my system at 100% and exercises a lot of memory, the RAM temperature used to get hot enough to fail one of the sticks. Fortunately, by doing some re-arrangement of the sticks to get better air flow over them, I was able to reduce the temperature to a point where this problem was resolved.

As for execution time decreasing with higher temperatures, there may be other reasons, such as cached data and such. Honestly, this would be a good question for Intel's chip design engineers. The apparent speed up could be as simple as clock skew when running hot. So, I guess that I am saying "I don't have a clue"... Anyway, I'll send a message to my old friend Bruce Ravenel who was one of the principal architects and engineers for the 8086/8087 processor family (though he left Intel back around 1980), and if he responds I'll let you know what he says. Alternatively, you could go to the Intel web site and post the question in their user forums. Their engineers often respond to questions like this.