After ~70% FPS Boost For Zink, The OpenGL-on-Vulkan Code Is ~50% The GL Native Speed
For those following the development of Zink as a software OpenGL driver built atop the Vulkan API, there are some new performance numbers to discuss this weekend.
Following the reports last week of 50~100% performance improvements for Zink by independent developer Mike Blumenkratz, he shared more details this week on his testing and the optimizations achieved.
As outlined a few days ago, he's been profiling and optimizing for the demanding Unigine Heaven OpenGL tech demo.
From where he started he is up ~70% compared to the prior state of Zink, which puts it at around 24 frames per second... 24 FPS isn't too exciting, but that's in the context of the notoriously demanding for its time Unigine Heaven.
With the hardware he's testing, what is more interesting is how it compares to a native OpenGL driver compared to going through Zink atop Vulkan. With his Intel system he is using for development and in turn having the Iris Gallium3D driver for the Intel graphics, he finds Zink to be running at around half the performance of that native driver. Previous to his recent optimization focus, it was around 25% the speed of Iris.
So Zink still nearly isn't as efficient as a native OpenGL hardware driver, but it's getting much better with time. It's only been recently that there has been a focus on performance after recently tackling OpenGL 4.6 support. I'll be running some Zink performance tests on my side soon, but for now those wanting to learn more about the latest Zink optimizations can see Mike's blog. Hopefully all of these patches will get reviewed timely and upstreamed into (ideally) Mesa 20.3.
Following the reports last week of 50~100% performance improvements for Zink by independent developer Mike Blumenkratz, he shared more details this week on his testing and the optimizations achieved.
As outlined a few days ago, he's been profiling and optimizing for the demanding Unigine Heaven OpenGL tech demo.
From where he started he is up ~70% compared to the prior state of Zink, which puts it at around 24 frames per second... 24 FPS isn't too exciting, but that's in the context of the notoriously demanding for its time Unigine Heaven.
With the hardware he's testing, what is more interesting is how it compares to a native OpenGL driver compared to going through Zink atop Vulkan. With his Intel system he is using for development and in turn having the Iris Gallium3D driver for the Intel graphics, he finds Zink to be running at around half the performance of that native driver. Previous to his recent optimization focus, it was around 25% the speed of Iris.
So Zink still nearly isn't as efficient as a native OpenGL hardware driver, but it's getting much better with time. It's only been recently that there has been a focus on performance after recently tackling OpenGL 4.6 support. I'll be running some Zink performance tests on my side soon, but for now those wanting to learn more about the latest Zink optimizations can see Mike's blog. Hopefully all of these patches will get reviewed timely and upstreamed into (ideally) Mesa 20.3.
43 Comments