AMDGPU-PRO 18.30 Pro/Open vs. Upstream Mesa OpenGL/Vulkan Radeon Benchmarks

[agd5f]

And what about opensourcing the userspace bits of the blob? Kinda like out-of-tree patches? Or would there be too much effort to adapt it to mesa (or what else might be necessary)?

IIRC, they were posted a while ago including mesa patches.

[bridgman]

And what about opensourcing the userspace bits of the blob? Kinda like out-of-tree patches? Or would there be too much effort to adapt it to mesa (or what else might be necessary)?

AFAIK if they are not already open sourced that's just a matter of finding time to do it. Believe the code is in the amdgpu X driver, will check.

[Termy]

The problem is that we can't get Freesync upstream until we have agreement from multiple vendors on a vendor-independent approach. That discussion does seem to be making progress but things like that rarely happen quickly.

And what about opensourcing the userspace bits of the blob? Kinda like out-of-tree patches? Or would there be too much effort to adapt it to mesa (or what else might be necessary)?

[bridgman]

i thought that should be sorted with ROCm and especially not that amdkfd will move into kernel?

The amdkfd driver has always been in the kernel - it was just built into a separate module from the graphics driver, partly so it could work with both radeon and amdgpu drivers and partly to serve as a relatively HW-independent reference driver for the HSA Foundation.

From a consumer side i'd wish for OSS Freesync-Support finally...

The problem is that we can't get Freesync upstream until we have agreement from multiple vendors on a vendor-independent approach. That discussion does seem to be making progress but things like that rarely happen quickly.

[fuzz]

Old versions of Mesa outperforming new versions is definitely not "how it should be".

It doesn't matter for the end user. Users shouldn't have to go out of their way to install new versions of their drivers. I do, but with this article I realized dealing with the extra PPA isn't worth it. Out of the box performance is extremely important. So yes, that is how it should be for the majority of users.

Developers have all of the information they need to deal with regressions.

[Enverex]

It surprised me too, but honestly that's how it should be!

Old versions of Mesa outperforming new versions is definitely not "how it should be".

So what's the conclusion? The results seem all over the place. Every stack wins in some games and loses in others.

[Termy]

I think that in OpenGl/Vulkan, AMD is pregressing in the right path, the way they are dealing with their statck.

Now, they need to put more attention, and support more AMD hardware in OpenCl, because in the server world its OpenCL/Cuda that rules...
No one is buying there, expensive cards to attach to a display...
Its for paralel processing, and in this world, Linux has a MAJOR Pie, and it tends to increase with time even more..

AMD has competive hardware, and in some Areas, it his even the best,
I see Rocm, with a big oportunity to AMD to gain steam, in GPU computing.

i thought that should be sorted with ROCm and especially not that amdkfd will move into kernel?

From a consumer side i'd wish for OSS Freesync-Support finally...

@topic: indeed nice to see that the stock experience is that good!

[tuxd3v]

I think that in OpenGl/Vulkan, AMD is pregressing in the right path, the way they are dealing with their statck.

Now, they need to put more attention, and support more AMD hardware in OpenCl, because in the server world its OpenCL/Cuda that rules...
No one is buying there, expensive cards to attach to a display...
Its for paralel processing, and in this world, Linux has a MAJOR Pie, and it tends to increase with time even more..

AMD has competive hardware, and in some Areas, it his even the best,
I see Rocm, with a big oportunity to AMD to gain steam, in GPU computing.

[Marc Driftmeyer]

Considering the Pro stack isn't using Mesa 18.3 what's the point of this comparison?