The ati petition

Vladimir Dergachev volodya at mindspring.com
Mon Sep 6 17:31:04 PDT 2004



On Mon, 6 Sep 2004, Mathieu Lacage wrote:

> On Mon, 2004-09-06 at 02:23, Vladimir Dergachev wrote:
>>>>> to the same power-performance tradeoff for your graphics card and saving
>>>>> power.
>>>> Good point. I would imagine that since most of GPU is inactive under
>>>> regular X drivers I am running at minimum power ;)
>>>
>>> Unlikely.  I suspect that, like with the CPU, the GPU speed must be
>>>  manually slowed.
>>
>> Unlike regular CPU, GPU consists of many blocks. I would expect that if 3d
>> engine is not used at all it consumes little power - kinda like running
>> HLT continuously on spare CPU.
>
> Actually, that is pretty unlikely. A lot of the power consumption of
> these devices (even when they are running at full speed) comes from the
> clock tree which distributes the clock throughout the chip. Unless the
> designers have added special multiplexers to stop completely the clock
> tree of the unused blocks, these will consume a lot of power even idle.

Possibly they did. There are registers in the documentation that deal with 
power management and they have bits like this:

    RE_CLK:     0   - Dynamic
                1   - Force on

To me this would mean that the RE (Render Engine) clock is not active if 
there is nothing for the render engine to do.

Keep in mind that this is mere speculation. Also, some ATI sample code 
writes 1 to such registers with comment that "some versions of ASIC do not
properly implement dynamic clocks". Linux drivers do not have this code 
though, so this might have been fixed in production version.

                         best

                           Vladimir Dergachev

> The basic clock rate used nowadays is high enough and the area of the
> chip die is big enough that these issues are non-negligeable. Of course,
> restarting the stopped blocks can also be a bit tricky for the chip
> designers.
>
> Another rather popular technique is to step the clock speed but that is
> much more complex to implement for the designers which is why few ASICs
> really do it.
>
> Of course, this is off-topic but I could not resist. Bad bad me. It
> should be said I have minimum experience in this domain so, the usual
> disclaimers apply.
>
> regards,
> Mathieu
> -- 
> Mathieu Lacage <mathieu at gnu.org>
>
>
> _______________________________________________
> xorg mailing list
> xorg at freedesktop.org
> http://freedesktop.org/mailman/listinfo/xorg
>



More information about the xorg mailing list