John Fremlin's blog: CPU throttling invalidating benchmarks

Posted 2010-03-15 23:00:00 GMT

I was trying to update the benchmarks for cl-irregsexp and to my surprise I found that as my laptop heated up it was getting slower. The observation was that the more times I repeated a benchmark, the slower it became.

I'd set the cpu_scaling_governors to performance, and made sure that the max scaling frequency was okay. But the CPUs were entering throttling states, as shown by acpitool.

One of the thermal zones was right on the throttle temperature when I looked at it

  Thermal zone 2 : ok, 85 C
  Trip points : 
  critical (S5):           130 C
  passive:                 85 C: tc1=0 tc2=3 tsp=10 devices=CPU0 CPU1 

The room I am in is unusually hot. I guess this may invalidate some of my benchmark results. Fortunately, people have repeated most of the HTTP ones to get similar numbers, but I shall have to run them again. With the heating off.

Does anybody know a way of checking whether throttling ever kicks in? I guess I shall poll the temperature every few seconds.

PS 20100318. Worked round it for long running benchmarks by forcing the CPU to be in the lowest possible speed, and moving to a colder room.

Post a comment