-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA512

Andre Müller wrote:
> Andre Müller wrote:
>> Peter Ganzhorn wrote:
>>> You may be glad to be wrong, today I managed to enable HPET on my ICH4
>>> chipset and it should be possible to get it running with ICH3 chipsets -
>>> AFAICT they do have hidden HPET functionalities.
>>> Just grab a 2.6.22-rc4 [1] kernel, patch it with 2.6.22-rc4-hrt2 [2] and
>>> there you go :)
>> i am glad I'm wrong, thank you!
> 
>> this really is fun. it compiles just fine,
>> and gives me: 1.4 wakeups per second on an idle console,
>> in c3 for 770 msec/each for 99.9 percent of the time,
>> processor down to 160 MHz. whoopsi.
>> ehci_hcd won't worsen things.
>> I did not stop even stop polling for network connection.
> 
>> in X (fluxbox):  8 wakeups, 150 msec/each in c3,
>> NB: I get no extra wakeups with opengl support via
>> the i915 in-kernel module.
>> with net.ppp0 : 13 wakeups, 130 ...
> 
>> ...and then thunderbird kicks in causing 11 timer calls per second
>> and driving wakeups to 30.
> 
>> but then: 37.4 was the lowest low before HPET.
> 
>> for reference, this is an IBM-R50e ICH-4M 1300MHz Celeron-M.
> 
> Update:
> suspend-to-ram will not resume anymore.
> it did not do so reliably befoer, so ...
> suspending to disk (software_1) looks good.

Another Update:
I did a bit of testing:
i get a patched firefox down to 25 wakeups per second,
resulting in a total of 50 wakeups in full-blown, but
idle, work environment.

Power consumption is at 11 Watts, .5 less than the
"classic" setup with a timer. Not percievably better
than a no-hpet setup, though.

Inactivating all of usb, net, sound, leaving acpid
for fan control gives 10.5 Watts, it was 11.2 on the
"classic" setup.

living on the console (on 1.7 wakeups/s) costs about .1W
less than fluxbox with opengl with as few as 6.2 wakeups.
And the fun bit: with screen blanking kicking in after
20min, i get to 7.3W -- 2 Watts less than the "classic",
before 2.6.21.

Without any acpi control (meaning the fan unceasingly
ventilates away for ever), 7.1W can be got. net in action,
it's 7.3W. This is all in X...

Finally, a word of warning: running the ondemand governor
on celeron-m (P4_CLOCKMOD) did nicely but ineficienty...
with hpet forced, the system every so often freezes at
switching performance -- very unstable.
cpufreq left alone, the system looks very stable, very friendly,
the processor popping out of c3 every now and then :-)

I propose a toast on Arjan and you, community,

Andre

_______________________________________________
Power mailing list
[email protected]
http://www.bughost.org/mailman/listinfo/power
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.7 (GNU/Linux)

iD8DBQFGaHvAhYOgL/M0VS0RCrVdAJ40/W+BxRIvQVi26o4g/gVaLzEpagCgky1Y
KSyPZAaKk/Jy+zNIs6D45Zo=
=lhxY
-----END PGP SIGNATURE-----
_______________________________________________
Power mailing list
[email protected]
http://www.bughost.org/mailman/listinfo/power

Reply via email to