AMD Catalyst vs. Radeon Gallium3D On Linux 3.1

[leif81]

Did you set it to low power profile?

How do you do that?

[marek]

The VDrift racing game results remain mixed, which is due to Mesa/Gallium3D not handling some of the shaders correctly and other incorrect renderings.

Wrong. It's because of Catalyst not handling some of the shaders correctly. The shaders are broken and shouldn't compile on a compliant driver. And you also need floating-point textures to render everything correctly, but I guess you already know that.

[mudig]

@Michael and others: How's the stability with the OSS radeon drivers recently? I'm asking because I'm using Fedora 15 and get random system hangs with my 5870 since the update to kernel 2.6.38.7 (https://bugzilla.redhat.com/show_bug.cgi?id=712533) which sucks HARD. Don't have any stability problems on my Samsung N140 netbook...

[pingufunkybeat]

I agree. I measured a few days ago my computer's power consumption, and I was shocked: The open source driver consumes 40W more than the official in idle (I have an 5770 in an arch linux machine).

Did you set it to low power profile?

I run all my cards in low (plenty for desktop performance) and switch to high performance only for games.

[bbandi]

I think it would be more relevant for the open source driver audience to review 2D and compositing.
Also in this case CPU/RAM/power consumption would be more important than raw performance.

I agree. I measured a few days ago my computer's power consumption, and I was shocked: The open source driver consumes 40W more than the official in idle (I have an 5770 in an arch linux machine).

[agd5f]

A new drm is not going to change 3D performance much. For 3D, the drm just takes the command buffer from the 3D driver and feeds it to the hw; not a lot has changed in that regard.

[FireBurn]

It would be interesting to see what the results would look like benchmarking the git head of mesa form the release dates of 2.6.39 3.0 & 3.1

As improvements happen in tandem to drm and mesa it would give a fairer picture of the improvements in the 3D stack

[Azpegath]

There's no point in continuously showing the VDrift performance results with the Gallium drivers if the rendering output is so horribly broken. That's equivalent to benchmarking AES encryption that spits out invalid data.

Perhaps update to the latest version of VDrift that was released a month ago?

[MatthiasF]

I think it would be more relevant for the open source driver audience to review 2D and compositing.
Also in this case CPU/RAM/power consumption would be more important than raw performance.

[gigaplex]

VDrift benchmarks are meaningless

There's no point in continuously showing the VDrift performance results with the Gallium drivers if the rendering output is so horribly broken. That's equivalent to benchmarking AES encryption that spits out invalid data.