Performance issue

[Roland Scheidegger]

Michael Stather wrote:
> OK, thanks for your help :)
> If this is not a bug and the performance is just not that good I can 
> live with that.
> Is the "radeon" driver still maintained and improved? I mean that some 
> features aren´t accelerated ATM, and I also heard that it uses just AGP 
> 1x because higher modes are buggy.
I assume you're really speaking about r200, not radeon (they share the 
same "radeon" 2d driver, but your 9250 uses the r200 3d driver). Plain 
old (r100) radeon would be almost full-featured, with the exception of 
3d textures and some ati env_bumpmap extension - there's a non-working 
patch for 3d textures, maybe one day I'll finish it (if the idea why 
it's non-working turns out to be true...) but don't count on it. As for 
r200, the "pixel" stuff is pretty full-featured too, it'll even support 
ati fragment shaders if you want. It lacks hw accelerated vertex 
programs unfortunately. Both of these drivers also are somewhat lacking 
wrt to large vertex arrays / vertex buffer objects currently, which can 
cause large performance drops for some apps (like ut2k3). Both drivers 
also lack support for occlusion queries.
AGP modes higher than 1x are not buggy per se, it's just that some 
chipsets have problems with it - this is not (graphic chip) driver 
specific though. Most motherboards IME should run just fine if you 
enable higher agp modes in your xorg config file (but don't expect too 
much performance improvements from that setting). This in contrast to 
agp fast writes which you can manually enable too, this has a pretty 
high probability of complete lockup at xorg start time (again, depends 
on the chipset), not to mention I think noone ever could show this would 
actually help performance...

Roland