Radeon RX 6800 "Sienna Cichlid" Firmware Added To Linux-Firmware.Git

[GrayShade]

bridgman sorry for the ping, but there are a couple of reports of 6800 XT not working on Linux at https://www.reddit.com/r/archlinux/c...rking_on_arch/. Do you think someone could look into it?

[Michael]

bridgman sorry for the ping, but there are a couple of reports of 6800 XT not working on Linux at https://www.reddit.com/r/archlinux/c...rking_on_arch/. Do you think someone could look into it?

Did any of them try Linux 5.10? As mentioned in the article, Linux 5.9 with some select Mesa state I encountered various AMDGPU errors but 5.10+ was fine in that case... So they may want to try 5.10 Git just to see if it works there.

[GrayShade]

Did any of them try Linux 5.10? As mentioned in the article, Linux 5.9 with some select Mesa state I encountered various AMDGPU errors but 5.10+ was fine in that case... So they may want to try 5.10 Git just to see if it works there.

I don't think so. I noticed that part in the review article, but you didn't mention if your issues were "game X crashed every so often" or rather "system got stuck at boot", so I assumed the former :-).

[Michael]

I don't think so. I noticed that part in the review article, but you didn't mention if your issues were "game X crashed every so often" or rather "system got stuck at boot", so I assumed the former :-).

Stuck display at boot, could SSH in and see AMDGPU errors. AMD later confirmed were tracking down the issue Mesa side.

[bridgman]

Stuck display at boot, could SSH in and see AMDGPU errors. AMD later confirmed were tracking down the issue Mesa side.

Our thinking at the time was that the problem was with Oibaf Mesa packages being built with an older llvm, but apparently that has been updated and i don't know whehter Michael picked up from Oibaf was before or after that.

Will go look at the Reddit page, thanks for the heads up.

[Michael]

Our thinking at the time was that the problem was with Oibaf Mesa packages being built with an older llvm, but apparently that has been updated and i don't know whehter Michael picked up from Oibaf was before or after that.

Will go look at the Reddit page, thanks for the heads up.

All my testing all along when using the Oibaf PPA has been with LLVM 11.

[bridgman]

All my testing all along when using the Oibaf PPA has been with LLVM 11.

Hmm, looks like we'll need to do more digging.

re: the reddit page I'm not sure yet was is causing the errors first poster is reporting (IB tests failing during kernel driver self test) but the problems on another page linked from that turned out to be missing numactl. I checked back with our packagers and we do have that listed as a dependency, so I gather either the dependency didn't get carried across to the Arch package or the missing dependency didn't get noticed during install.

[Guest]

Nice. Obviously would be perfect to do it day earlier. But this is still speedy.

This isn't speedy, this is very late. Intel finishes this shit months ago. And the GPU driver is unusable without these firmware files. This is once again lste support by AMD.

But the trend indicates that they're improving. Hopefully they keep improving and get to Intel's level some day.

[Guest]

Maybe or part of Wayland stack on GPU at least.





Yes but

https://www.phoronix.com/scan.php?page=news_item&px=GCC-For-AMD-GCN-Early

and

Porting GCC to AMD GCN microarchitecture
Honza Hubička SuSE ČR s.r.oPrague
Joint work with Martin Jambor
GNU Cauldron 2016

Slide 4

https://gcc.gnu.org/wiki/cauldron2016?action=AttachFile&do=view&target=Porting+GCC+to+GCN.pdf

And I mention

RDNA / RDNA2 is less similar to CPU then GCN
CDNA is more similar to CPU then GCN

So I it may be advantage-

And If- then is slowed down on CPUs due to Spectre/Metldown and other Transient Execution Attacks

https://arxiv.org/pdf/1811.05441.pdf

and on GPU you are able to run both/all possible branches of code and use only valid branch result to speed up code on GPU

It's not the Wayland protocol that you need to run on the GPU. Mainly just render as much as possible on the GPU, do blitting and compositing on the GPU and don't keep transferring buffers back and forth between the GPU for each frame. And especially the stupidity of writing to individual pixels from CPU.

A one way pipeline from CPU rendering ------> GPU rendering ------> presentation is best for performance.

[GrayShade]

re: the reddit page I'm not sure yet was is causing the errors first poster is reporting (IB tests failing during kernel driver self test)

I can confirm that mine works fine with 5.10, but not 5.9.