The Major Open-Source ATI Improvements Over Two Years

[marek]

Do you know when these things will get 'ported' to r600g? Or are they not needed / supported by the hardware?

The r600 hardware can do it, but the code to get it running is very hardware-specific. Some ideas are the same, but very little (if any) code can be shared. I guess we will have to write it from scratch.

[69acid69]

Hi,
what's the current situation/state with "GL_ARB_depth_clamp" extension on r300g, because it's disabled?

Thanks.

[marek]

Hi,
what's the current situation/state with "GL_ARB_depth_clamp" extension on r300g, because it's disabled?

Thanks.

We weren't able to make it work.

[tball]

The r600 hardware can do it, but the code to get it running is very hardware-specific. Some ideas are the same, but very little (if any) code can be shared. I guess we will have to write it from scratch.

Thanks for the reply.
How big a performance improvement do you think z-compression and hyperz will give?

[Kano]

I tested it with openarena, disabled swapbufferswait for full performace. With my x1600 256 mb i think there is a vram limit. I got an impressive speed increase from 62 to 84 fps @ 800x600, but with higher res the speed even decreased a tiny bit from 43 to 42 fps @ 1024x768 and 29 to 28 fps @ 1280x1024. So if you use the RADEON_HYPERZ=1 override use a benchmark to verify. I dont have got a faster card with more vram to do a better test. It was definitely not cpu limited with my i7-880.

[yotambien]

I tested it with openarena, disabled swapbufferswait for full performace. With my x1600 256 mb i think there is a vram limit. I got an impressive speed increase from 62 to 84 fps @ 800x600, but with higher res the speed even decreased a tiny bit from 43 to 42 fps @ 1024x768 and 29 to 28 fps @ 1280x1024. So if you use the RADEON_HYPERZ=1 override use a benchmark to verify. I dont have got a faster card with more vram to do a better test. It was definitely not cpu limited with my i7-880.

How do you disable vsync? I thought it was hard coded in the driver. Or better yet, what is the proper way to benchmark OA? Some time ago I downloaded that anholt demo but it didn't run here...

[Kano]

Look there:

http://kanotix.com/files/fix/oss-test/readme.txt

In that dir you find scripts which patch the driver (or enable gallium) when you add any option. I used pts to benchmark openarena. Maybe i begin a new profile next time.

[yotambien]

Ah, I downloaded pts and now I see how the OA benchmark is done. I'll have to compile the latest drivers, it's been a while for I was quite happy with the last one.

[yotambien]

I tested it with openarena, disabled swapbufferswait for full performace. With my x1600 256 mb i think there is a vram limit. I got an impressive speed increase from 62 to 84 fps @ 800x600, but with higher res the speed even decreased a tiny bit from 43 to 42 fps @ 1024x768 and 29 to 28 fps @ 1280x1024. So if you use the RADEON_HYPERZ=1 override use a benchmark to verify. I dont have got a faster card with more vram to do a better test. It was definitely not cpu limited with my i7-880.

9600M, mesa 7.10, swapbufferwait 0, vblank_mode 0, hyperz 0->1

800x600: 65.4 -> 72.8 (11%)
1024x768: 48.4 -> 56.9 (17%)

This was locking the CPU to its maximum frequency. If I don't do that the frame rate is worse, but the improvement due to hyperz is more noticeable (56.9 to 65.7 at 800x600, 15% better).

Very nice indeed!

[Kano]

Did you do

dmesg|grep nouveau|grep core

to see if your gpu is running at full speed? i dont think so, looks so slow. ask in irc://irc.freenode.net/#nouveau then when c (current) is not the max.