Don't Expect AMDGPU To Enable SI/CIK Support By Default Anytime Soon

[phoronix]

Phoronix: Don't Expect AMDGPU To Enable SI/CIK Support By Default Anytime Soon

It seems every few days a discussion among end-users and Linux gamers re-emerge about their belief that the CIK (GCN 1.1) support any even the newer SI (GCN 1.0) support should be enabled by default in the AMDGPU kernel driver to succeed the Radeon DRM driver...

http://www.phoronix.com/scan.php?page=news_item&px=AMDGPU-CI-SIK-No-Default-Soon

[juno]

I agree with Michel Dänzer here, as long as there is no audio support, it should not be default. It's not that big deal right now with this state of radv. However, things get tricky in case there will finally be free OpenCL and Vulkan support on top of amdgpu while DAL features are still missing.

[funfunctor]

If folks want GCN 1 cards as defaulted to amdgpu they first need to talk to their distro vendor. This change needs to come from the bottom up, once the ecosystem is happy with the right versions of libdrm, mesa, llvm and so on its its being used in this way in the wild then and only then are we likely to toggle it over upstream.

[Decay]

So, how is the DAL stack rolling along? Could it be ready for mainlining anytime soon?

[boxie]

So, how is the DAL stack rolling along? Could it be ready for mainlining anytime soon?

in the open source kernel. it will be a while. from the talk that was given by AMD people at a recent talk - there was a bunch of work that needed to get done and AMD people are time poor

That and differing ideas with kernel people on what should be done

[schmidtbag]

Though I completely understand the issue revolving around which driver to pick, I would argue that's inevitably a non-issue. To my understanding, all GPUs in radeonSI should be compatible with what amdgpu is trying to do. If the GCN 1.0 and 1.1 GPUs end up being fully functional in amdgpu, then doesn't radeonsi effectively become obsolete? Distros could just simply replace the old drivers, or treat their radeonsi package as a meta-package.

[Guest]

schmidtbag I think they meant that the AMDGPU and Radeon kernel drivers would conflict with each other. AFAIK Radeon SI mesa driver is used for newer hardware as well.

[bug77]

How much nvidia pays to phoronix to write partially false news against amd. This and Deus Ex with amd graphics has similar non pro writing style. No news that gtx 1080 has burned etc. Poor Nouveau driver is more important.

In a custom amd-staging-4.7 kernel, enable support for SI, CIK, DAL, whatever you need.

You can only fry a GTX 1080 if you play Furmark for hours. Since Furmark is not a Linux application, I don't see the relevance in a Linux forum. Oh, I almost forgot: you need to get that 1080 from EVGA and on top of that you need to pick a particular model (FTW).
The problem with AMDGPU not supporting older cards or Deus Ex not working on AMD cards is a bit more generic than that, don't you agree?

[bug77]

Support for older amd cards is there like it is for nvidia card, with nvidia you need compile dkms kernel driver, with amd enable support in the kernel.

Those look like very different approaches to me (compile kernel driver vs compile kernel).

A lot of games works with amd graphics so the problem with deus ex are in game producer company using optional bindless plaaplaah function for example.

The article on Phoronix was not about whose problem it is, but rather that if you have an AMD video card, you can't play DX:MD. Arguably, you can't play it if you have a Nvidia card either, looking at performance numbers, but that's a different discussion.

Bottom line, you're trolling as usual, posting about unrelated subjects in a thread about support for older cards in AMDGPU.

[bridgman]

Though I completely understand the issue revolving around which driver to pick, I would argue that's inevitably a non-issue. To my understanding, all GPUs in radeonSI should be compatible with what amdgpu is trying to do.

If the GCN 1.0 and 1.1 GPUs end up being fully functional in amdgpu, then doesn't radeonsi effectively become obsolete? Distros could just simply replace the old drivers, or treat their radeonsi package as a meta-package.

The radeonsi driver is a userspace Gallium3D pipe driver that supports GL, CL and multimedia on GCN parts. The amdgpu driver under discussion here is a kernel driver.

If you mean "doesn't radeon (kernel driver) become obsolete ?" then sort-of yes but only for GCN parts. The radeon kernel driver would still be required for the other 9 GPU generations it supports, ie everything from r100 through NI. It would just no longer be required to support SI and CIK.