The AMD Radeon Performance Is Incredible On Linux 3.12

[lano1106]

I do not understand Alex possible explanation concerning GART. I thought that this feature has disappear long ago. With the last system having GART are over 10 years old.

The only thing that I am aware is the i915 driver that have a fake GART driver. Is the radeon driver is using a similar trick?

[pingufunkybeat]

Comparing the catalyst results for HD 6950 from this article, the OSS driver is as fast as or faster than Catalyst BUT in the reaction quake test.

Are you sure about that? It seems to be between 60-90% on average on 6950.

[ChrisXY]

Somehow I get the impression that cheap gpus run well and expensive ones run bad.

I tried to use the open source driver for my hybrid HD 7970M.

Auto poweroff ("runtime pm") does not seem to be in 3.12 yet. It's pretty important because with vgaswitcheroo if you power off the radeon gpu in X and then go to a tty and then back to X, the driver locks up.

With DPM enabled:
When starting X on boot it locked up with a cursor in the upper left corner. Starting X a little while after boot worked.
When powering off the radeon GPU with vgaswitcheroo X locks up completely.

That's not even performance or features, that's the most basic stuff like not locking up X.

[Kostas]

So.. has erroneous rendering been completely ruled out?

[Michael]

So.. has erroneous rendering been completely ruled out?

Everything looks the same to me when I watch all the tests.

[oleid]

Are you sure about that? It seems to be between 60-90% on average on 6950.

I mixed up the the results of radeon for low screen resolution at OpenArena 0.8.8 with Catalyst's results for high resolution when I wrote my comment above. But apart from that, it looks quite good for radeon. Now, tests of more demanding games would be interesting.

Open Arena 0.8.5, 1920x1080:
3.11: 450 93%
3.12: 455 94%
Cata: 482 100%

Open Arena 0.8.8, 1920x1080:
3.11: 101 46%
3.12: 116 53%
Cata: 216 100%

Reaction Quake, 1920x1080:
3.11: 184 24%
3.12: 344 57%
Cata: 599 100%

Xotonic 0.7, 1920x1080, Low:
3.11: 317 121%
3.12: 343 130%
Cata: 263 100%

Xotonic 0.7, 1920x1080, High:
3.11: 183 97%
3.12: 221 118%
Cata: 188 100%

Xotonic 0.7, 1920x1080, Ultra:
3.11: 161 100%
3.12: 190 119%
Cata: 161 100%

Edit: If you sum up the FPS and devide them by Catalyst's results, you get:

3.11: 1396 73%
3.12: 1669 87%
Cata: 1909 100%

[mark45]

I disabled Adblock to click on some ads but unfortunately it seems the university's proxy blocks the ads too. Willing or not it seems I'm not going to see the (in)famous phoronix ads

I disabled Adblock but there are animated flash banners which make the CPU fan spin quicker and make more noise so I enabled it back, without animation I'd turn adblock off for phoronix.

[YAFU]

Hello.
I have read that AMD is finally making some progress in the OpenCL proprietary implementation to can work with Blender 3D and other similar programs. Is AMD working on improving OpenSource OpenCL implementation too?

[przemoli]

Hello.
I have read that AMD is finally making some progress in the OpenCL proprietary implementation to can work with Blender 3D and other similar programs. Is AMD working on improving OpenSource OpenCL implementation too?

Yes.

(Though Gallium OpenCL is behind. But it may have better compiler which is such a pain for Blender...)

[M1kkko]

With most games being distributed via Steam, see: http://www.phoronix.com/scan.php?pag...tem&px=MTQxMzY

For Wine games or others, new PTS test profiles are welcome and would be glad to run them.

Michael,

Every now and then when a change comes up which could substantially affect linux gaming performance, we need to see some current-generation games/benchmarks being used for testing. Knowing there has been some progress in benchmarking with Steam / Source engine games, perhaps such benchmarks should be seen every once in a while. And whenever it becomes too tiring to use Source engine for benchmarking, Unigine Valley would provide a pretty good alternative.