Zink OpenGL-On-Vulkan Now "100%-1000% Faster" For Many Scenarios

[oleid]

Zink is the future, everyone will use it.

I do, it is great for corrosion prevention!

[camel_case]

Zink is the future, everyone will use it.

[smitty3268]

Zink really seems to have the potential to get within striking distance or practically at parity with native OpenGL drivers.

I think that remains to be seen. This is good news, but I suspect it's still likely to be significantly slower than native in a bunch of tests.

Eventually, it makes one wonder if Gallium3D will even remain useful

Zink runs on Gallium3D, so if you think Zink is useful then so is Gallium3D.

Even some older generation hardware such as Terascale (supported by the r600 driver) could theoretically have a Vulkan driver developed for it, which might ease support for such older beasts in the longer term.

Seems very unlikely, and since such hardware is no longer changing creating a whole new driver is going to be significantly more work than just finishing the OpenGL one and then never touching it again.

[SteamPunker]

Zink really seems to have the potential to get within striking distance or practically at parity with native OpenGL drivers. In some cases, this was already true with Zink even before this latest peformance breakthrough. Once it gets even close to native OpenGL driver performance, driver maintainers and Mesa developers should really consider deprecating native OpenGL drivers and concentrate on making the best possible Vulkan drivers, and assisting with further development of Zink. If Mike Blumenkrantz managed to get Zink to be this good from scratch, all by himself, imagine how much it can be improved even further, once a lot of driver development get freed up due to no longer having to maintain separate hardware-specific OpenGL drivers?

Eventually, it makes one wonder if Gallium3D will even remain useful, since Vulkan can perfectly play that same lower layer closer-to-the-metal role as well. Gallium3D will only continue to have value for legacy hardware that lack the features necessary to support Vulkan. Even some older generation hardware such as Terascale (supported by the r600 driver) could theoretically have a Vulkan driver developed for it, which might ease support for such older beasts in the longer term.

And when it comes to newer generation GPUs still to be released, the manufacturers and driver developers would be crazy to waste any more resources on developing Vulkan drivers and OpenGL drivers in parallel. Just focus entirely on Vulkan drivers, and let Zink handle the OpenGL stuff..

[R41N3R]

Well, my last test with Tomb Raider felt already strange on Mesa 21.1 and I couldn't believe the numbers I got with my RadeonVI in 4K.
RadeonSI: min 88, max 180, avg 154
Zink: min 134, max 231, avg 194 (edit: I think lightning like e.g. the sun is not rendered correctly)
DXVK: min 113, max 190, avg 161

It's great to see how well Zink is developing and I hope there will be more benchmarks soon.

[Quackdoc]

I assume compared to the old Zink code because I don't think it would be possible with an increase of that much performance of the native OpenGL drivers which should be fairly optimized by now.

Some drivers are some aren't I assume some will get major benefits Raspberry pi zink could be really big once it's vulkan driver supports it, I know my baytrail might benefit, as it's zink preformance is about 50% so it may even get close to native. also Virtio-gpu will see some massive gains from this I believe.

But Yes I believe he would be referring to comparison to previous zink. most drivers should be getting close to comparable though.

[onlyLinuxLuvUBack]

got phoronix?

Im Ready++++++++++++++++++++++++++++++++++++++++++

^future tshirt

[uid313]

By 100-1000% faster he means compared to the old zink code or to native GL drivers?

I assume compared to the old Zink code because I don't think it would be possible with an increase of that much performance of the native OpenGL drivers which should be fairly optimized by now.

[tildearrow]

By 100-1000% faster he means compared to the old zink code or to native GL drivers?

Most likely to the old Zink code.

[Quackdoc]

Oh boy, that is a massive jump, cannot wait to test this out. will be cool if crappy old laptops actually get a preformance jump because of this.

EDIT: I assume chrome OS's linux VM will be using zink by default, I would love to see the benchmarks.