A Fresh Look At The AMD Radeon Gallium3D Performance

[phoronix]

Phoronix: A Fresh Look At The AMD Radeon Gallium3D Performance

As noted earlier in the week, the open-source AMD Radeon "R600g" driver that supports 3D acceleration on Radeon HD 2000 series graphics cards through the latest Radeon HD 6000 and Fusion graphics processors, is becoming quite fit. The driver is nearing a point of stability, is mature enough for most desktop users, and it is beginning to receive some performance optimizations and other improvements. Thanks to this recent work, plus the ongoing development of the Linux 3.0 kernel, here is a fresh set of AMD Gallium3D Linux driver benchmarks.

http://www.phoronix.com/vr.php?view=16110

[siride]

Can you include things like KWin and Compiz performance? Those are still pretty important, though as things keep getting faster, they will end up being non-interesting after a few years.

[Kano]

This is a much better benchmark comparison than the ones before. Disabling SwapbuffersWait has usally the most impact of benchmarks. I did not test the impact of the other option used. The problem with SwapbuffersWait is when a game runs in the 20-25 fps range, then tearing is really extreme, like Trine, in that case i usually enable it. But for benchmarking you definitely should turn it off. Maybe test with installelled libtxc_dxtn and enabled texture-float unigine heaven, but of course in that case you need to disable tesselation in the benchmark.

[Tgui]

Excellent news and excellent article!

[smitty3268]

This is a much better benchmark comparison than the ones before. Disabling SwapbuffersWait has usally the most impact of benchmarks. I did not test the impact of the other option used. The problem with SwapbuffersWait is when a game runs in the 20-25 fps range, then tearing is really extreme, like Trine, in that case i usually enable it. But for benchmarking you definitely should turn it off. Maybe test with installelled libtxc_dxtn and enabled texture-float unigine heaven, but of course in that case you need to disable tesselation in the benchmark.

Agreed. These benchmarks with everything over 100 fps are of questionable use, but overall this was a much better set of tests than usual. I really wish Michael would benchmark some more challenging apps, like Doom3, Portal, etc. in addition to the standard Quake 3 engines.

[89c51]

would it be easy for someone to explain why we don't have consistent performance across resolutions ?? (at least Catalyst seems to fluctuate less than the Open solutions)

[Michael]

I really wish Michael would benchmark some more challenging apps

Most still don't run well on Mesa / G3D.... For those that do, submit a Phoronix Test Suite test profile for them if you want to see them used and then they get picked up right away for use in articles.

[EvilTwin]

how about benching the blender view-port performance in different modes?
this is something you actually use compared to benches like lightsmark

[Michael]

how about benching the blender view-port performance in different modes?
this is something you actually use compared to benches like lightsmark

Submit a test profile and it will be included...

[mirza]

Ok, now we are getting somewhere....