Zink Mesa 21.3-dev Benchmarks - Increasingly Capable Of Running OpenGL Games Atop Vulkan

[phoronix]

Phoronix: Zink Mesa 21.3-dev Benchmarks - Increasingly Capable Of Running OpenGL Games Atop Vulkan

Zink as an OpenGL-over-Vulkan API implementation living within Mesa merged its sub-allocator code that could deliver 10x the performance for some games. Plus it also landed OpenGL compatibility context support for getting more games working now with this open-source GL-on-VLK solution. Given the progress made in Mesa Git over the past week, here are some fresh benchmarks now for how the performance stands across various games and benchmarks.

https://www.phoronix.com/vr.php?view=30480

[kbios]

Thanks Michael, very interesting benchmark! The xonotic result is particularly promising, showing that there is not necessarily a performance loss using a translation layer

[strtj]

"code that could deliver 10x the performance for some games" - wake me up when anything even close to this actually happens. I barely saw 2x on the most favorable real-world benchmark.

[curfew]

I hope these benchmarks put the baseless Zinc hype down for a while. Any modern game that puts a decent load on the CPU as well will suffer of a noticeable overhead with OpenGL-to-Vulkan translation.

[Tawheed]

Wunder how zync will perform under cad workload. Specviewperf is heavily optimized by quadro and firepro drivers while the standard drivers lack 20x in performance ( siemens nx).

[MadByte]

Thanks for the benchmarks!
It's impressive how much it improved, surely not 10x better for most cases, but even one game running faster via zink then OGL is really cool to see.
So, great work Mike! Can't wait to see more of it.

[oibaf]

Impressive improvements indeed!
Hopefully tesseract performance regression will be fixed in the future.

[curfew]

Thanks for the benchmarks!
It's impressive how much it improved, surely not 10x better for most cases, but even one game running faster via zink then OGL is really cool to see.
So, great work Mike! Can't wait to see more of it.

Rather it begs the question why is that particular game running so slow with native OpenGL. The drivers are not perfect and there will be bugs and bottlenecks, especially with OpenGL 4 features. (I faintly recall that OpenGL 3 was deemed acceptable a few years ago while anything above that was universally broken.)

[microcode]

It's cool watching the edge cases melt away week by week.

[Black_Fox]

Thanks for the benchmarks! I started the article with low gaming FPS expectations and was very pleasantly surprised. Then I found the additional lazy descriptors results at the end... feels like that Vince McMahon meme.