Zink Suballocator Lands In Mesa - "Over 1000%" Performance Increase For Some Games

[phoronix]

Phoronix: Zink Suballocator Lands In Mesa - "Over 1000%" Performance Increase For Some Games

Mesa's Zink Gallium3D code for implementing OpenGL-over-Vulkan can now run a heck of a lot faster with the newest Mesa 21.3 code...

https://www.phoronix.com/scan.php?page=news_item&px=Zink-Suballocator-Merged

[cewbdex]

Would be nice to have benchmarks of some applications and games

[tildearrow]

Would be nice to have benchmarks of some applications and games

He said that in the article.

I'll be firing up some new Zink benchmarks shortly.

[Michael]

He said that in the article.

Yep already started them.

[kiffmet]

Since Zink is still built around Gallium3D paradigms, wouldn't it theoretically be possible to run D3D9 (or DX10/11) games via Nine (or a hypothetical DX10/11 state tracker) and have GPU handed the work via the Zink=>Vulkan pathway? I understand that we have fantastic alternatives to this approach with DXVK and VKD3D-Proton, but from a nerd perspective I am wondering how these solutions would compare against each other if fully developed.

[oiaohm]

Since Zink is still built around Gallium3D paradigms, wouldn't it theoretically be possible to run D3D9 (or DX10/11) games via Nine (or a hypothetical DX10/11 state tracker) and have GPU handed the work via the Zink=>Vulkan pathway? I understand that we have fantastic alternatives to this approach with DXVK and VKD3D-Proton, but from a nerd perspective I am wondering how these solutions would compare against each other if fully developed.

https://gitlab.freedesktop.org/mesa/mesa/-/issues/3170

That not quite as straight forwards as it first sounds. One day it may work. The reality here is that Nine part is not exactly on the same set of Gallium3d Paradigms as Zink. Zink was built using the newer Paradigms. There is a lot of update work to Nine to make it use newer gallium3d Paradigms. Yes without this work there will be extra overhead in translation from old Gallium3d Paradigms to the newer ones.

Gallium3d Paradigms are not a 100 percent static constant they do change over time.

[Quackdoc]

Since Zink is still built around Gallium3D paradigms, wouldn't it theoretically be possible to run D3D9 (or DX10/11) games via Nine (or a hypothetical DX10/11 state tracker) and have GPU handed the work via the Zink=>Vulkan pathway? I understand that we have fantastic alternatives to this approach with DXVK and VKD3D-Proton, but from a nerd perspective I am wondering how these solutions would compare against each other if fully developed.

yes you can run g9 over zink I believe, I havent tested it since I borked my install, but it should work to some degree now. in theory the current d3d10umd should also work on zink, but it may need a lot of work to get wired up.

[Azrael5]

An extraordinary result. impressive. Linux development is able to optimize hardware in extraordinary way.

[Adarion]

Ah, we shall celebrate these times when big progress is still possible.
In a few years, maybe, people will hardly be gainig more than some 0.x percent performance. On the other hand, things are hopefully matured then.

Each of these patches brings Linux based operating systems more towards mainstream suitability and a better experience for already existing users.

[You-]

An extraordinary result. impressive. Linux development is able to optimize hardware in extraordinary way.

Extraordinary, but without benckmarks and comparisons we wont know how extraordinary.

If it takes Vulkan performance in theplaces impacted to above 50% of native, impressive. If it takes it from 0.5% to 5% then its still a big stepping stone, but a stepping stone.

(My calculation being that in the long term we will accept old normally incompatible software being 5x faster on 10x faster hardware. But if we get closer to native or supercede it, that is all a nice bonus.)