Vulkan vs. OpenGL Performance For Counter-Strike: Global Offensive

[phoronix]

Phoronix: Vulkan vs. OpenGL Performance For Counter-Strike: Global Offensive

Valve yesterday introduced experimental Vulkan support for Counter-Strike: Global Offensive as an alternative to the long-standing OpenGL path for running CS:GO on Linux. Curious about the performance implications of CS:GO with Vulkan, here are some benchmarks on Ubuntu Linux with a variety of AMD Radeon and NVIDIA GeForce graphics cards.

https://www.phoronix.com/vr.php?view=30762

[tildearrow]

I wonder, are the HLSL shaders compiled to SPIR-V at run-time or pre-compiled?

[MastaG]

I wonder, are the HLSL shaders compiled to SPIR-V at run-time or pre-compiled?

I believe they are compiled at shutdown but also as a weekly cron job..

[tildearrow]

I believe they are compiled at shutdown but also as a weekly cron job..

Come on, I was just trying to find the source of stuttering...

[aufkrawall]

Don't seem to be precompiled yet. Start bot matches and fly through maps as spectator to accumulate state cache.
Once that is done, frame time variance seems to be much better than with GL in Danger Zone mode.

[BNieuwenhuizen]

Michael Are you sure the right resolution is used for dxvk csgo? I hear some people can't even test on lower res and on the lower end cards all your dxvk results look suspiciously like the 4k results.

[fimmy]

I saw yesterday's article and was keen to try the new -vulkan launch option. My enthusiasm didn't last long however when my character would run 15 feet and the the game would drop 50 frames, rinse and repeat. I also wasn't able to zoom around the map when dead like I normally could. It would change to a black screen after a few short moments of zooming. Im using a 2080Super on a Ryzen 5900X. There has been a driver update overnight 495.46-2 so I'm hoping to try it out again hopefully with better results.

[Dudi]

CS:GO is Source (1). Not Source 2. Source 2 has native Vulkan and doesn't need to rely on DXVK-Native.

[fimmy]

looks like the overnight driver update has indeed fixed yesterdays issues.

[skeevy420]

If it was CPU bottlenecked then there wouldn't be framerate differences across GPUs.

That assumes that all the GPU drivers are fully optimized. Also, Nvidia GPUs are using vanilla 5.13 with no PPA and AMD GPUs are using 5.16 and Oibaf PPA so kernel & mesa differences could be a factor. This isn't a direct comparison between brands.

For fairness sake I think AMDGPU(-Pro) with the same kernel and Mesa used as Nvidia should have been benchmarked.