A few questions about integrated graphics

[Chris Pemberton]

Rui Tiago Cação Matos wrote:
> On 18/08/2008, Chris Pemberton <cjpembo.mailing.lists at gmail.com> wrote:
>   
>>  4.  Any mesa/intel xorg driver code "optimized" for 64-bit, CPU cache
>>  size, multiple cores, etc...?
>>
>>  5.  Some software (I think mplayer is one and most certainly ATLAS)
>>  ignore your compiler settings and create optimized code automagically.
>>  Anything I can do when compiling xorg/intel/mesa drivers to improve the
>>  performance?
>>     
>
> AFAIK, graphics chips don't understand x86 intructions nor x86_64
> though they have registers where you put commands and/or a command
> buffer. But these commands are issued dinamically by the graphics
> driver and not by your compiler so there isn't much gcc can do aside
> from generating more efficient driver code (the one that actually runs
> on your CPU and issues those commands).
>
> I think there is work going on about using more dynamic "graphics
> commands synthesizers" like LLVM in graphics drivers[1].
>
> Rui
>
> [1] http://zrusin.blogspot.com/2007/05/mesa-and-llvm.html
>
>   
Thanks you for the explanation and link:  llvm seems both interesting 
and promising.

My last question can best be asked using a specific example:

An Intel DG33TL motherboard has a GMA 3100 graphics processor built in.  
It can accept CPUs with a 1333 or 1066 or 800 MHz system bus, and either 
667 or 800 MHz RAM.  Obviously if any graphics processing is offloaded 
to the CPU, clock speed, bus speed and ram speed are crucial.  But for 
graphic processing handled solely by the integrated GPU, will the bus 
speed or ram speed have any effect?  Or does the GPU communicate with 
the shared RAM at a "fixed" speed beyond the control of the system 
builder?   Did that make sense?  Will choosing the faster system bus or 
ram speed allow the integrated GPU to read/write to shared memory more 
efficiently?

Thanks,

Chris