Re: [Dri-devel] PCI gart: possible fix?

[Alex D. <ag...@ya...>]

--- Michel Dänzer <mi...@da...> wrote:
> On Wed, 2004-02-11 at 21:26, Alex Deucher wrote:
> > I was looking through the the databooks regarding PCI gart and I
> > noticed that r128, radeon, and r200 all do PCI gart the same way. 
> that
> > said, I also noticed that PCI GART is limited to 4 MB of system
> memory
> > when used with a PCI card and 32 MB when used with an AGP card.  
> 
> Where did you see that?

It's in the r128 DDK software development guide, page 2-23.

> 
> > I haven't yet looked at the gart code in the radeon driver, but
> perhaps
> > it is trying to use more than 4 MB of system memory 
> 
> The default is 8 MB for both AGP and PCI GART.
> 
> > and that is causing the slow down.  
> 
> Perhaps, but I'd rather expect different symptoms, and it used to
> work
> fine for me on older chips. Easy enough to find out though:
> 
> 	Option	"AGPSize" "4"

Yeah, it's just a guess.  I'm not familiar with the code, nor do I have
a PCI card to test with.

> 
> ("GARTSize" is an alias in newer radeon drivers)
> 
> > it may also explain why PCI gart seems to "work" on AGP cards and
> older 
> > radeons, but not on newer PCI ones. 
> 
> >From what I remember, mostly AGP cards used to be problematic
> traditionally, and only on x86. PCI cards being problematic is a
> recent
> trend.

Yeah.  As I recall it seems like mostly r200 PCI cards.

Alex

> 
> 
> -- 
> Earthling Michel Dänzer      |     Debian (powerpc), X and DRI
> developer
> Libre software enthusiast    |  
> http://svcs.affero.net/rm.php?r=daenzer
> 
>

__________________________________
Do you Yahoo!?
Yahoo! Finance: Get your refund fast by filing online.
http://taxes.yahoo.com/filing.html