Re: [tlug] CPU speed: reported vs. actual

[Darren Cook <darren@example.com>]

>  > I tried installing and using cpufreq-set as Mattia suggested, with no
>  > improvement; next thing to try is booting into Windows and running
>  > Intel's linpack and some other benchmarks from there.

I finally got around to doing this. Those of you who were waiting with
bated breath can fill your lungs again.

Under Windows the machine got a CPU mark of 3870.2, which appears to be
consistent with other people with the same CPU. CPU usage was at 100%
(across all 8 virtual cores) during this.

I also ran the linpack benchmark. What was interesting was CPU usage was
at 100% for the quick "1000" run, but was at exactly 50% for the much
longer running "20,000" run. It gave me 12.604 Gflops (for the 20,000
run), i.e. roughly the same as the 11.647 I got under linux (I'm sure I
have more background processes running under linux).
This was very interesting as from watching "top" it seemed the linux
version was also doing this. This would explain why I get just under
half the 27 GFLOPS that Intel report [1].

Current Conclusion: Intel's own linpack benchmark is broken on an
i7-740QM CPU, or they use an algorithm that cannot use more than 4
threads. (BTW, it reports it needs 3GB for the 20,000 test, and I have
8GB; I'm running the 64-bit version.)

With that cleared up to my satisfaction, I think I can go back to where
I started (the subject line, in fact) and say that the dynamic CPU speed
switching (under linux) is not working: when it says 933Mhz it is not
really saving power but is still running at 1.7Ghz.

Darren




-- 
Darren Cook, Software Researcher/Developer

http://dcook.org/work/ (About me and my work)
http://dcook.org/blogs.html (My blogs and articles)