RADV+Zink vs. RadeonSI OpenGL Performance On Mesa 23.2-devel

[Michael]

Heard from Mike that he's been already able to reproduce and fix some of the bigger gaps

[rmfx]

Heard from Mike that he's been already able to reproduce and fix some of the bigger gaps

Cool, would be nice to see this benchmark v1.1

Maybe including a bit more games would be interesting since it's quite synthetic test here.

[schmidtbag]

I'm a little confused why Zink ever outperforms a native driver. With DXVK it was a little more obvious because it would basically make things multi-threaded, but Mesa can be multi-threaded, so where's the performance boost coming from?

[Michael]

Maybe including a bit more games would be interesting since it's quite synthetic test here.

I'm always up for including more benchmarks but limited by the ones that meet my standards around automation/scripting support... I think the only OpenGL games I left out are even older titles or not running on modern distros. If there is any automated-friendly games I missed out on, please let me know.

[Quackdoc]

I'm always up for including more benchmarks but limited by the ones that meet my standards around automation/scripting support... I think the only OpenGL games I left out are even older titles or not running on modern distros. If there is any automated-friendly games I missed out on, please let me know.

how do you do the automation? things like input scripting compatible?

[Michael]

how do you do the automation? things like input scripting compatible?

It all depends on the game.. My test automation scripts @ https://github.com/phoronix-test-sui...ree/master/pts

For the most part with games relying upon whatever functionality is exposed by the game itself due to other solutions being unreliable for reproducibility, not working e.g. Waylsnd vs. X.Org, or other limitations.

[oiaohm]

I'm a little confused why Zink ever outperforms a native driver. With DXVK it was a little more obvious because it would basically make things multi-threaded, but Mesa can be multi-threaded, so where's the performance boost coming from?

Some of it is simple enough.

https://mesamatrix.net/

Go down to "Extensions that are not part of any OpenGL or OpenGL ES version" and notice that Zink has 9 more optional opengl extensions than radeonSI has. Zink is the most complete opengl implementation mesa3d has. So zink can allow programs to go code paths they would not be able to when using radeonSI and those code paths can be better performing.

Yes there is still 9 more extensions documented Khronos Group that Zink does not have yet.
57 extensions at Khronos
48 extension by zink.
39 extensions by radeonSI

The 9 difference between zink and radeonSi those 9 exist as implemented with Intel/Nvidia/Qualcomm drivers of course as random-ally spread.. So programs have code paths for when they have intel/Nvidia/qualcomm that cannot run with radeonSi because opengl extension does not exist but when you use Zink can run.

Yes some of those extensions were design to make opengl operations less complex.

Yes zink implements more of the opengl than Nvidia closed source drivers as well.

The other thing here is when these benchmarks happen you normally see zink lead developer look at stuff and fix stuff and you also see radeonSI drivers look at stuff and fix stuff. For a long time radeonSI really did not have solid competition inside mesa.

[oiaohm]

And in particular, when Zink + RADV is consistently close or beating RadeonSI, it'll maybe open the discussion for eventually ditching all the special Gallium3D drivers in favor of Zink when all of the supported targets have adequate Vulkan drivers..

Remember Zink is a gallium3d driver. Also that going to be awhile. There are still GPU being made that don't have the functionality to run vulkan but to have the functionality to implement some amount of opengl. Also there is the case as well on some of these embedded GPU where their vulkan implementation equals zink explosing less opengl than what the native opengl driver can.

That eventually ditching all opengl Gallium3d drivers other than zink that could be a few decades out before that is fully practical. But it also possible that opengl only GPU in embedded usage stay around forever due to cheap licensing.

[dragorth]

Fantastic; it is especially promising seeing Zink win on some of these tasks, even more so to see it so close on some of the Unigine marks (which are absolute GPU monsters).
And in particular, when Zink + RADV is consistently close or beating RadeonSI, it'll maybe open the discussion for eventually ditching all the special Gallium3D drivers in favor of Zink when all of the supported targets have adequate Vulkan drivers... at which point Gallium3D can be refactored into Zink itself, and supporting OpenGL well will be a matter of tuning Zink and implementing Vulkan extensions. :+ )

Zink relies on Gallium3D, and Zink will want to have things to compare to so this probably won't happen like that.

[QwertyChouskie]

I'm always up for including more benchmarks but limited by the ones that meet my standards around automation/scripting support... I think the only OpenGL games I left out are even older titles or not running on modern distros. If there is any automated-friendly games I missed out on, please let me know.

SuperTuxKart is quite automation-friendly. Everything can be set via command-line flags. There should even be a PTS profile for it, but it may need to be updated to the latest version of STK (1.4).