[ltp] T61 temperature limits

[Owen Heisler]

--pE2VAHO2njSJCslu
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
Content-Transfer-Encoding: quoted-printable

On Thu, 2008.04.17, 108, Jeffrey Taylor wrote:
> Quoting Henrique de Moraes Holschuh <hmh@hmh.eng.br>:
> > Nice alone can't do that.  You want a process to use less power?  Let it
> > end (or sleep) as quickly as possible so that the CPU can go to C3
> > ("race to idle").  If it is a long running process where "race to idle"
> > doesn't matter (i.e. it doesn't sleep, for example), you either stop it
> > for a while and then let it run again for another while (using
> > SIGSTOP/SIGCONT for example), or you use nice and tell the cpu speed
> > governor to ignore niced loads.
> >=20
> > SIGSTOP+SIGCONT will probably waste less power, but you'd have to
> > measure it to know, as that depends on the workload and also on just how
> > much time in C3 your CPU will stay when the background process is not
> > running.
>=20
> There is a program to run a program, STOPing and CONTinuing with variable
> weighting.  The project is at:
> http://sourceforge.net/projects/dutycycle
>=20
> Source tarball available for download.  It does reduce the temperature.  =
On my
> laptop, kernel builds boost the temperature from around 40C to around 70C.
> With dutycycle on the default 50%, temperatures run 55C-60C.

Isn't this what throttling does?

http://acpi.sourceforge.net/documentation/processor.html (legacy docs)
 Throttling is sort of enforced power management: Even when the system is=
=20
 highly active, the CPU is "put to sleep" for short amounts of time.

http://en.wikipedia.org/wiki/CPU_throttling
 According to the ACPI Specs, the C0 working state of a modern day CPU can =
be=20
 divided into the so called "P"-states (performance states) which allow clo=
ck=20
 rate reduction and "T"-states (throttling states) which will further throt=
tle=20
 down a CPU (but not the actual clock rate) by inserting STPCLK (stop clock=
)=20
 signals and thus omitting duty cycles.

According to those two excerpts, using a T state that throttles 50% should=
=20
force the processor(s) to spend at least half the clock cycles in a sleep=
=20
state.

If that is true, then the most effective way to reduce both heat and power=
=20
usage is to throttle the processor but use the ondemand cpufreq governor to=
=20
control the processor's frequency, considering that the processor is more=
=20
efficient at a higher clock speed:
http://www.thinkwiki.org/wiki/How_to_reduce_power_consumption#CPU
http://www.bughost.org/pipermail/power/2007-May/000166.html

But maybe it is not true, as suggested by this on thinkwiki.org.  Maybe=20
throttling a processor does not force it to drop to a sleep state.  Why not?
http://www.thinkwiki.org/wiki/How_to_make_use_of_Dynamic_Frequency_Scaling#=
A_note_about_CPU_throttling

--pE2VAHO2njSJCslu
Content-Type: application/pgp-signature; name="signature.asc"
Content-Description: Digital signature
Content-Disposition: inline

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.6 (GNU/Linux)

iD8DBQFIB6O99XC8Sh1vpNARAhobAJ91jXPlft2I0DP3TMU+zGPQhO9TgQCfQDWu
62TkBhnpZ1B+wXRH/NAHx+c=
=Odd+
-----END PGP SIGNATURE-----

--pE2VAHO2njSJCslu--