Open-Source Radeon Vulkan Driver Now Can Run Dota 2 With Correct Rendering

[DrYak]

Well, I get ~84 fps with pts on a R9 285 (~87 fps with the experimental RADV_HIZ_DISABLE=0), compared to ~90 fps with mesa, so I think we're past the 80% on this game already. The remaining difference is going to get harder though as the whole state management wrt compressed surfaces is kind of a mess currently.

Thanks and congrats for the awesome work.

Second, if the driver is passing conformance tests, why the games are having problems? That means that the problem is with a bad implementation of Vulkan in the games?

Or simply games stressing parts that aren't covered by current conformance test? (i.e.: the games are NOT non-conformant, the games simply hit corner cases nobody though of before).

[BNieuwenhuizen]

Second, if the driver is passing conformance tests, why the games are having problems? That means that the problem is with a bad implementation of Vulkan in the games? There is conformance tests for games/applications?

The results of the conformance test suite that were posted on the livejournal blog are
[114598/114598] skip: 50388, pass: 62932, fail: 1193, timeout: 2, crash: 83

In particular, we pass some of the tests, not all of them. I think the phoronix wording is very misleading here.

Furthermore, a test suite can't tests every possbility & corner case in the implementation so even if we pass everything we could still have bugs. The test suite also has some vulkan features which are not tested at all. For example, vkQuake uses input attachments these days, which are a required vulkan feature, but the test suite doesn't test them. Neither anv nor radv support input attachments.

[pal666]

Second, if the driver is passing conformance tests, why the games are having problems?

it is unrealistic to tests everything

[pal666]

btw, the clickbait blog title is misleading, the correct answer should be 'no'

[haagch]

It also seems that the polaris regression happened due to one of the commits that gave a significant performance boost, so we might need to do a step back for a while.

If anyone is interested: https://github.com/airlied/mesa/comm...3c814b12726a05

Well, it doesn't *need* to mean much. After all this was only tested with drm-next-4.9-wip, so there's a chance it's really a polaris specific kernel problem since it works on older GPUs generations...

[dungeon]

Well, I get ~84 fps with pts on a R9 285 (~87 fps with the experimental RADV_HIZ_DISABLE=0), compared to ~90 fps with mesa, so I think we're past the 80% on this game already. The remaining difference is going to get harder though as the whole state management wrt compressed surfaces is kind of a mess currently.

I usualy meant percentages so 80% of the pro blob, but OK

Speaking about blob... VK Dota2 is much faster then GL, while VK Talos currently is sort of the same as blob GL.

[raytracer]

Amazing effort guys! I personally cannot wait (although I will patiently follow the progress) when I can compile vkQuake and fire it up on my laptop with SI chipset

[Guest]

This is great progress! I look forward to trying it out sometime in the future.

[darkbasic]

Cool, thanks alot Dave!

@Michael
Please test Pro Vulkan vs FOSS Vulkan on Dota 2. I know it's still early code, but it would be interesting anyway. Thanks

[Michael]

Cool, thanks alot Dave!

@Michael
Please test Pro Vulkan vs FOSS Vulkan on Dota 2. I know it's still early code, but it would be interesting anyway. Thanks

Yes that's obviously part of the plan aside from OpenGL results too.