May 2011: Gallium3D vs. Classic Mesa vs. Catalyst

[Azpegath]

The Catalyst driver didn't work with Warzow on the 4670?

[Azpegath]

These benchmarks look fishy. Last time I tried openarena anholt demo my fps was much higher on HD 5750.

Are you running with swapbufferswait on again? It's likely to have a negative effect on fps. You can argue all you want about defaults being used, but with this setting on it's unlikely you'll see much performance progress being made in some benchmarks, making the benchmarks kind of pointless.

Insanity: doing the same thing over and over again and expecting different results -- Albert Einstein

From the article:

The results are not very different at 1920 x 1080. Of course, if Swapbufferwait and other non-standard tweaks were made, additional performance could be tuned out of the open-source driver stack, but still it would be no match to Catalyst.

[LinuxID10T]

These benchmarks look fishy. Last time I tried openarena anholt demo my fps was much higher on HD 5750.

Are you running with swapbufferswait on again? It's likely to have a negative effect on fps. You can argue all you want about defaults being used, but with this setting on it's unlikely you'll see much performance progress being made in some benchmarks, making the benchmarks kind of pointless.

Insanity: doing the same thing over and over again and expecting different results -- Albert Einstein

I would definitely agree with you on this one. I have a HD 5750 as well and I can tell you it will easily play Nexuiz with decent settings at HD resolutions. Hell, even my Core i3's integrated graphics will do that. Something is definitely off about these tests.

[darkbasic]

Another useless phoronix benchmark. Nobody should ever use vsync for benchmarks

[smitty3268]

Since the code that was tested, Tom Stellard committed his r300g register allocator changes.

And it would be nice if Phoronix could update the tests used to add some that don't go over 150fps on Catalyst, to see if the CPU limitations of the r600g really hurt when it matters or not, and how much. I know that all kinds of more complicated games work on the driver. Really anything other than a Quake3 based game would be nice. I've heard Unigine Heaven even works, although you couldn't compare it to Catalyst since the advanced rendering features would be disabled.

[smitty3268]

Oh, and whatever happened

to that promise that we'd get tables listing the hardware and software used? By that, did you mean you would link to OpenBenchmarking and still just list everything in text form on Phoronix? It seems like these promises of "it's coming" were made about a year ago by now. And it would only take 15 seconds to fix this yourself - just open up a spreadsheet and copy paste a screenshot, for goodness sakes.

[elanthis]

yes i wanna see unigine heaven 2.5. just deactivate everything r600 cannot render.

That would be just about everything. Literally, I don't think Heaven can run on the Mesa drivers at all, period.

There were some commits a couple weeks ago adding features that would bring Mesa _closer_ to running Heaven, but not actually making it possible.

[allquixotic]

Look on the bright side: Gallium3d is faster than classic now, and in most tests, the evergreen card was actually faster (or at least on par) with previous-generation cards. Wow! That never happened before (memories of r300g kicking r600g/evergreen's butt in every benchmark...)

Unfortunately I'm stuck on mesa 7.10.2 until someone plugs in an r600 card to a system with a COM port (mine doesn't have a COM port nor even a header) to get a kernel backtrace to figure out why it panics while running Imprudence.

[Kano]

Would usb to serial work too?

[smitty3268]

and in most tests, the evergreen card was actually faster (or at least on par) with previous-generation cards. Wow! That never happened before (memories of r300g kicking r600g/evergreen's butt in every benchmark...)

Dave Arlie committed some code to r600g on April 19th (the day before this test) which stopped flushing everything to the GPU all the time, and i think that gave a pretty significant speed boost.