As far as I understood in the blog, he isn't really working on performance yet, and the mentioned gains are the by-product of a long debugging session.
Announcement
Collapse
No announcement yet.
Zink OpenGL-On-Vulkan Seeing Some 50~100% FPS Gains
Collapse
X
-
Not every OpenGL application would necessarily benefit from Zink, if they have certain pipeline stalls (e.g. GL software running on Wine that also do GDI overlays, frame buffer readback, CPU heavy immediate mode shennanigans, as Autodesk products do).
But most games would probably benefit because they typically have their own decoupled render loop, which if bottlenecked only by GPU throughput, Vulkan-style driver-side parallelism & reduced API overhead could potentially produce some significant performance gains.
- Likes 3
Comment
-
Originally posted by Remdul View PostNot every OpenGL application would necessarily benefit from Zink, if they have certain pipeline stalls (e.g. GL software running on Wine that also do GDI overlays, frame buffer readback, CPU heavy immediate mode shennanigans, as Autodesk products do).
But most games would probably benefit because they typically have their own decoupled render loop, which if bottlenecked only by GPU throughput, Vulkan-style driver-side parallelism & reduced API overhead could potentially produce some significant performance gains.
- Likes 2
Comment
-
Originally posted by EmbraceUnity View PostWhat were the previous benchmarks? If Zink used to be three times slower, then now it is only 50% slower. If it used to be 50% slower, now it is up to 25% faster.
If it was 50% slower before, then "50%-100% faster" means 75% to 100% the speed of OpenGL
0.5 x 1.5 = 0.75
0.5 x 2.0 = 1.0
- Likes 1
Comment
-
Originally posted by smitty3268 View Post
The previous post was not actual performance, so this is not in comparison to that. That was mostly concerned with shader compile speed, not runtime.
I do think it would be interesting to get some actual benchmark numbers from Michael. A bunch of people seem to believe Zink is fast for some reason, so it'd be good to get actual confirmation on that - or that, as expected, it's still really slow.
The only questions are, what we can have right now, and how far we are from the goal.
- Likes 3
Comment
-
Originally posted by zxy_thf View PostConsider DirectX-over-Vulkan cab be faster than the native implementation sometimes, there is no reason why OpenGL-over-Vulcan can't have comparable performance.
The only questions are, what we can have right now, and how far we are from the goal.
Buzzwords are nice, but just putting "vulkan" in the title of something doesn't magically make it fast.
Comment
-
Originally posted by smitty3268 View Post
There are absolutely reasons to believe that. Check out this very blog post from the article for an example, where he talks about less than ideal handling of samplers due to the mismatch between the vulkan and GL rules.
Buzzwords are nice, but just putting "vulkan" in the title of something doesn't magically make it fast.
Comment
Comment