Re: [gpm] About GPU speed handling in g-p-m

[Richard Hughes <hughsient gmail com>]

On Sun, 2006-11-26 at 14:49 +0100, benjamin canou wrote:
>   Hi,

Adding in nvclock maintainer...

> I think it may be interesting for g-p-m to handle GPU speed.
> After having discussed a bit about this with Richard, I launch this
> thread to start collecting informations and thinking about an
> implementation, as well as performing tests to verify the real impact on
> power consumption.

Yes, we need to work out if it's worth the extra work. I've a sneaky
feeling my GPU is pretty power hungry, but we need to be able to do this
in a safe way. See below.

> Here are some links for NVIDIA cards:
> 
> There is the nvclock tool:
> http://www.linuxhardware.org/nvclock/
> There is also a thing called CoolBits for the "official" support, there
> are some hints in the FAQ:
> http://download.nvidia.com/XFree86/Linux-x86/1.0-9629/README/chapter-04.html

Yup, we need to work out a simple safe way to do this on 100% supported
cards, for instance:

nvclock -p [0-100] Sets the power consumption by changing the memory and
core frequencies. 0 is minimum permissible speed, and 100 is maximum
speed. Overclocking is not allowed in this mode, and if the card isn't
supported (like mine isn't)[1] then the command fails.

We can then add support into HAL for this, and mask on VID/PID's - doing
it this way lets us use the same method name (but with a different tool)
for Intel and ATI.

Then we can do a tickybox "Reduce video performance when on battery [x]"
in g-p-m.

Richard.

[1] my video card is a Quadro NVS 110M / GeForce Go 7300 - and nvclock
doesn't get the core frequency correct... PID=0x1d7, VID=0x10de