ATI Linux Drivers

OldMan wrote:
I bought a 7500 ayear ago just cause I tought it would get linux drivers earlier…and they skiped it… ggrrrrrr… I got furious with that. Now I have a Useless card and no money to buy another.

Then wrote:
Are the current ATI driver a completely separated stuff from DRI ones?

Yes.

Did ATI anounced any plan of integrarting them at Xfree distributions?

No.

I would be really mad if I buy a r9700 and discover its drivers produces lots of artifacts like the 7500 does.

Haven’t you learned yet? Don’t buy a video card unless there’s Free (open source) drivers for it. Hmm… the status page at http://dri.sourceforge.net/ lists your card as supported. Are you using the DRI drivers currently?

Originally posted by OldMan:
If anyone has a 9700 runnign on linux… some screenshots would be nice.

My "Infinite Terrain" demo running in Linux

The fruit of porting my framework to Linux makes it’s appearance.

cool!

I `ve learned… I wil buy only NVIDIA from now… my GF 2 GTS runs almost 4 times faster than the r7500 in linux

Originally posted by Humus:
[b] My "Infinite Terrain" demo running in Linux

The fruit of porting my framework to Linux makes it’s appearance. [/b]

I am proud that this appearance came in answer to one of my questions…

But I would like to know something else Humus… what is your opinion about these drivers. Are they SERIOUS stuff? Good performance? Or just something to make linux user stop crying? Too much slower than windows drivers?

ATI MAY have a chance in my next buy if these drivers are really serious stuff (i needed to re-think my last post after I saw GeForce FX expected price).

Concerning Performance: I’m getting very mixed readings here. I’ve been running a bunch of benchmarks over the last couple days, and the results are all but conclusive.

I’ve compared two effectively identical machines (Dual Athlon 1900), one with a GF4 (drivers 3123), one with the Radeon 9700.

On viewperf I get very similar results. The Radeon is a lot faster in UGS, which is not surprising as that test is heavy in AA lines, which are artifically slow on the GF4.

In machtest the Radeon is mostly faster (sometimes by a factor of 2), but slower in some tests.

On my own apps (which are admittedly not heavily optimized) the Radeon is twice as fast as the GF4. So it seems the Radeon is nicer for non-optimized code.

But the really interesting test is GLperf. The results for the Radeon are pretty strange, as it seems to oscillate a lot for longer primitives. I’ve put some preliminary results at http://www.igd.fhg.de/~reiners/radeon, will run some more tonight.

Has anybody tried running similar benchmarks on their system? What are the results?

Thanks

Dirk

Originally posted by OldMan:
[b]
I am proud that this appearance came in answer to one of my questions…

But I would like to know something else Humus… what is your opinion about these drivers. Are they SERIOUS stuff? Good performance? Or just something to make linux user stop crying? Too much slower than windows drivers?

ATI MAY have a chance in my next buy if these drivers are really serious stuff (i needed to re-think my last post after I saw GeForce FX expected price).[/b]

These driver works pretty well. I haven’t done any serious tests, but performance seams to be roughly the same as the windows driver. Out of the 6 demos I have tried to port to windows 5 works fine, the last works but after a while it exits with a messeage that it hit a breakpoint. As I don’t have any breakpoints in my code (I don’t even know how to do that with g++ ) I assume it hits some breakpoint in the driver. Will try to investigate that more deeply.
The issues I have found:
Anisotropic filtering doesn’t look quite right, it causes some aliasing. It’s as if they boosted the LOD too.
There seams to be a problem with legacy support, some old extensions that have been added to the core API are exposed, but when you try to get the function pointers you get either NULL or an invalid pointer. Grabbing “glTexImage3DEXT” gives you an invalid pointer for instance, grabbing “glTexImage3D” works fine though.

Otherwise it seams to work just fine.

The Linux demos are now available
http://humus2.campus.luth.se/~humus/

Really thank you! So it seems that are serious drivers. good… now i can have an alternative to NVIDIA. I really wanted this so I could keep a NVIDIA card at one PC and an ATI one at other.

Can’t wait to try out those demos, but I can’t access the page…

Bah

Originally posted by dirk:
[b]

Has anybody tried running similar benchmarks on their system? What are the results?

[/b]

I will try to run them in the 7500… lets see if it can perform good enought to apear at result ploting

Originally posted by richardve:
[b]Can’t wait to try out those demos, but I can’t access the page…

Bah [/b]

Try again. The whole area network went down not long after I posted this and was down the whole night. It’s up again now though.

Ah, it works again…

But now I’m having another problem; I’m getting a segmentation fault when running your demos.

It’s crashing inside the NVIDIA (binary) driver when calling glXCreateContext(…).

Probably an invalid pointer or so… I’ll fix it when I’ve got some time (in a few hours)

Oh, and some specs:

GeForce3 Ti 200
Latest Linux Detonators (so that’s not 40.xx)
Red Hat 8.0
XFree 4.2.0
Linux kernel 2.4/2.5 (I’m using both, rather 2.5 because it feels more snappy/polished)

Hm, okay… tracked down the problem…

I’m running my desktop in 16 bpp (MPlayer crashes when using 32 bpp) and it seems that glXChooseVisual() does return a valid visual when using 4 for the GLX_RED/GREEN/BLUE_SIZE and 16 for GLX_DEPTH_SIZE.

It looks like this (probably also because of the 16 bpp) http://home.quicknet.nl/qn/prive/rm.vaneijbergen/hgl.png

Well, I haven’t finetuned my app for checking invalid parameters and such yet. The VolumetricFogging demo requires an alpha channel, so it needs to run in 32bit mode, which also explains why it looks like in that screenshot.

I am still fighting with some linking errors here. Also noticed that it does not compile in gcc under 3.1x…

I see. I have only tried it with gcc 3.2, heh, but people should upgrade anyway

Oh, btw… Humus, could you please chmod the executable bits away from all of your source files?

Heh, yeah
I’ll do that the next time I update them. Had to pack it in Linux, but had some trouble getting it right, and finally I forgot to fix the flags.

So where does one get this new ATI driver?