Hi,

On Thu, Aug 16, 2007 at 10:04:24PM +0200, Erik Andrén wrote:
> All,
> 
> I've been experimenting on enabling the hidden hpet on an ICH3-based 
> system (Dell Latitude C640 P4 Mobile) using Thomas Gleixners patches 
> [1]. My results are based on patching the Ubuntu 2.6.22-9 kernel.
> 
> Power usage (in Watt)
> Kernel               Idle    Tweaked Idle    Full CPU Load
> 2.6.20-15-generic    34      -               60
> 2.6.22-ubuntu        34      32              60
> 2.6.22-ubuntu-hpet   34      32              60
> 
> As you can see from the result, no net gain is to be seen at all.

You don't mention any idle-state residing statistics at all.
The way I see it, HPET *only* helps if the longer *maximum idle
period* that HPET allows for (i.e. longer maximum timeout than most
other timer hardware!) can be successfully used to prolong idle state,
i.e. the system is already so idle that there are frequent wakeups
due to less suitable timers with much smaller maximum timeout intervals
(which causes unwanted wakeups).

I.e. if your idle times (ACPI Cx residency) are so short that the system
decides to remain stuck at promoting down to C2 only despite the system
actually featuring support down to C4, there's no gain in power use.

One should pay much more attention to what kind of residency length is achieved
and which residency would be required to reach the next deeper Cx
state, since that one would probably save power then.
(do you reach lowest Cx in /proc/acpi/processor/CPU0/power ??)

Oh, and it might well be that your system cannot be brought lower
than what you already achieved.

These are thoughts from someone who's not entirely well-trained in these
issues however, so it might not be entirely accurate, and from someone
who's full of envy due to having an Inspiron 8000 with HPET-incapable ICH2,
only one generation before your ICH3 HPET chipset.

HTH,

Andreas Mohr

_______________________________________________
Power mailing list
[email protected]
http://www.bughost.org/mailman/listinfo/power

Reply via email to