AMD's Open-Source Mesa Driver Continues To Be Ruthlessly Optimized For Workstation Performance

[aufkrawall]

Also, Intels Linux drivers are just a much better experience than the windows ones. They might lack performance in some cases, but in regards to weird bugs is far superior.

I wonder if Intel Mesa and Windows use the same shader compiler.

[ezst036]

AMD has a weird strategy with multiple redundant stacks. They should just open source everything.

It makes sense to me. A company has a piece of software that for whatever reasons(legal or otherwise) they simply can't open source, so they develop a whole new stack that from its inception has as its goal to be an open source stack, and over time work to continue to deprecate the use cases for the closed stack in favor of the open stack until the open stack is superior in every way.

[khnazile]

Vega gpus with dedicated memory are still crashing randomly: https://gitlab.freedesktop.org/drm/amd/-/issues/716
There's no sign of AMD doing anything about that. This is very disappointing, and that is one of the reasons why I can't recommend buying their products. Opensource drivers have a very little value when they don't work.

[PublicNuisance]

So it took 10+ years from playing complete catchup, to getting to the point "Binary blob might be completely forgotten about". Good job everyone involved.
A pleasure to be an AMD linux customer.

Not exactly, firmware blobs will still be present and required.

[aufkrawall]

I have random reboots on Windows too with ASPM enabled, be it with RDNA2 or Ampere GPU. It is not like proprietary drivers would miraculously not be affected by such (platform) bugs.
amdgpu kernel driver + Mesa proved to be very stable here...

[user1]

For the last few months AMD Windows drivers had frequent driver timeouts (black screen for a few seconds) on Polaris products, which still aren't fixed in the latest driver. They require a restart because they sometimes happen almost every minute. I've used Windows a little bit not long ago and tbh, on Linux everything feels much more stable because it currently has zero issues with Polaris (at least in my experience).

[Lech]

Vega gpus with dedicated memory are still crashing randomly: https://gitlab.freedesktop.org/drm/amd/-/issues/716
There's no sign of AMD doing anything about that. This is very disappointing, and that is one of the reasons why I can't recommend buying their products. Opensource drivers have a very little value when they don't work.

Yep, can confirm on vega 56. I'm not sure if they will ever fix it, it's really annoying even if the crashes are rare.

[ferry]

"The current gap for Snx [Siemens NX] between pro drivers and Mesa is ~2% with my private branch."

Strange, according to wikipedia "Starting with version 1847, support for Windows versions prior to Windows 10 as well as for macOS was completely removed, and the GUI was removed from the Linux version."

So, why is opengl performance measured with Snx?

[Teggs]

So, why is opengl performance measured with Snx?

Because the poster Marek was replying to referenced Snx explicitly.

[Viki Ai]

This is why I use Radeon GPUs at home. And at work where I can (to the point of having IT just buy me a Radeon Card instead of upgrading my whole machine since Dell presently is only shipping Nvidia-based systems, at least via what our IT department offers - I am less fussy on machines that will only be using Windows, but this particular box is going to be running Linux for GPU-heavy applications and life is too short and my time too valuable to be messing about with closed Nvidia drivers!)