Mesa Zink Improvements For OpenGL-On-Vulkan Reportedly Make It Faster Than Radeon OpenGL

[pinguinpc]

I meant on Windows. The Intel ARC driver is currently very buggy and slow with D3D9/11 but works fine with Vulkan.

Windows is meh, arc must be better on linux but only if cards stay avalaible in most countries

[brucethemoose]

I wonder if this would help Minecraft? Among "OpenGL apps that need to run faster," its the 10,000lb gorilla.

I am working on a Python script to bench it, but getting the client to run reliably is particularly frustrating. For instance, it likes to just randomly and silently fail to start up.

[fagnerln]

I would be interested in this too. However, I'm not sure if CSGO can run on DXVK. My quick test today failed with some errors about connecting to steam client.

If you use -vulkan as launch options, it will use dxvk-native

[shmerl]

Does zink support all kind of settings that help readeonsi work around quirks in games?

[smitty3268]

smitty3268 there is nothing click-bait on the title nor the article, because Zink was quite close to RadeonSI since May, as Michael already pointed out.

In that occasion, Zink with RADV was about 86% the speed of the RadeonSI driver on the RX 6800 XT, also being faster in a few cases. Of course it was slower in many ones, but now you know that it's not only in one micro-benchmark, as you affirm.

The author of Zink itself has said you should never expect Zink to outperform radeonsi, but nobody wants to hear that.

There's zero reason to believe that Zink outperforms radeonsi in large numbers of tests, and a huge number of reasons to believe it doesn't. How about the fact that this commit specifically calls out 1 micro-benchmark as being faster, rather than, say, an actual game or full demo application. Or even listing 2 micro-benchmarks. It doesn't, because this change won't magically speed up the whole driver by 20% across real applications. It speeds up a micro-benchmark that might then affect real apps by 1%.

The ideal is for Zink to come close enough to native drivers that nobody really cares if it's a bit slower, because it's close enough to not matter. It's getting close to that point already, at least in some applications. We need some more exhaustive testing to make sure that's the case across the board, and I hope Michael does said tests.

[Eirikr1848]

I wonder if this would help Minecraft? Among "OpenGL apps that need to run faster," its the 10,000lb gorilla.

I am working on a Python script to bench it, but getting the client to run reliably is particularly frustrating. For instance, it likes to just randomly and silently fail to start up.

Hopefully once you have it working, it could become part of the Phoronix Test Suite for automated benchmarking.

Minecraft benching would be a great way to test OpenGL on Vulkan performance for sure.

[theriddick]

If anyone is running benchmarks, maybe turn off your 60hz vsync limit

[ET3D]

That's one heck of a click-bait title by Michael.

Agreed. The entire article looks like it's reading too much into one microbenchmark. However, if Michael's enthusiasm for this performance improvement is making him benchmark Zink again, I'm certainly not complaining. Let's see how far Zink has come in the past few months.

[brent]

This is just a single microbenchmark. It's an impressive feat, but this probably doesn't mean much for applications or games.

[oibaf]

Zink has surely a higher overhead than plain OpenGL, but with radeonsi cards has the advantage of leveraging aco compliler in radv, while radeonsi still uses old llvm compiler, and this may be an advantage in some cases.