The Open-Source AMD Linux Driver Stack Hitting Problems With The Radeon RX 590

[grok]

Since this is 12nm maybe UVD block is also is changed/updated

Maybe it supports VC9 I guess not, but maybe yes who knows

It is quite possible that nothing changed at all. Some of this was seen of the Ryzen 2 series (Zen+) where they kept the same "floor plan" for everything except most of the stuff was a bit smaller. They fixed some very small detail perhaps at just one of the layers (I don't know enough about these matters to be more precise), added more sensors (temperature, power) to match how the power management is on the Zen APU.

So we do expect the Polaris 30, RX590 to be virtually identical to the Polaris 20 / RX580 and kind of an "optical shrink".
I think of nvidia GPUs ported a decade ago, from 65nm to 55nm, I don't think anything changed then? besides the performance and efficiency. So were they had a Geforce GTX 280, they released a GTX 285 instead.


-------

So there is this little linux surprise with the 590. Maybe the graphics board has changes (outside the GPU), new power components and such, GPU firmware a bit different, and tuned closed to the last minute for the market conditions. e.g. beating the nvidia GTX 1060 with GDDR5X in bar graph length.
I suppose, tiny little things throw the initialisation out of sequence and the driver just very easily fails to work.

Probably there is much the same going on with Radeon on POWER9 motherboards (Talos). Should be simple to recommend the RX480 or RX580, or the RX570 since it is again extremely similar but if it is untested and there is some tiny mismatch then this might or should make the driver fail to run.
This is interesting to learn about.
In this case there should be a very, very small number of people trying an RX480 on a POWER9 workstation motherboard (small motherboard runs to begin with but most users might use it as a test bed for the servers or to build software, and use the simple ASPEED graphics and remote shells. other users will get the "out of the box" package with the Radeon WX7100 instead)

[TemplarGR]

Maybe a firmware issue?I am not familiar with the inner workings of amdgpu, but it could be a possibility. After all, it is just a Polaris refresh, it should just work.

[Beherit]

Michael,

I have similar issues with my Special Edition Sapphire RX580 8 GB card when using the 4.19 kernel. dmesg reports ring failures and cannot recognize the minimum clock speed. I can use 3d acceleration and set proper resolution, but have several other issues making this kernel unusable for me.

Downgrading to 4.18.xx gets rid of all problems for me and no amdgpu errors appear in the dmesg log.

[dungeon]

It is quite possible that nothing changed at all.

Well for the driver every asic is different

other users will get the "out of the box" package with the Radeon WX7100 instead

These are probably safest to recommend, as these are Built by ATI and goes all the same... OK AMD.

Powered by ATI is different story, these customize everything, clocks are different, BIOSes are different, etc... If you say just RX 590 to someone, you can't be 100% sure, as that is different from vendor to vendor.

[perpetually high]

Michael,

I have similar issues with my Special Edition Sapphire RX580 8 GB card when using the 4.19 kernel. dmesg reports ring failures and cannot recognize the minimum clock speed. I can use 3d acceleration and set proper resolution, but have several other issues making this kernel unusable for me.

Downgrading to 4.18.xx gets rid of all problems for me and no amdgpu errors appear in the dmesg log.

It’s a real shame how bad that 4.19 LTS kernel is for AMD gpus. Hoping the fixes make it in eventually. 4.18 has been rock solid.

[dungeon]

Michael,

I have similar issues with my Special Edition Sapphire RX580 8 GB card when using the 4.19 kernel. dmesg reports ring failures and cannot recognize the minimum clock speed. I can use 3d acceleration and set proper resolution, but have several other issues making this kernel unusable for me.

Downgrading to 4.18.xx gets rid of all problems for me and no amdgpu errors appear in the dmesg log.

Ah, so we now have Two Sapphires reported not behaving right in harassment free for everybody way

[Guest]

That's the exact same problem I faced with my R9 390, except it's the VCE that fails to initialize. In my case, lm_sensors fails to work due to this.

[dungeon]

Oh even Grenada, such a harassment free for everybody island had an Operation Urgent Fury

[Beherit]

dungeon, I've no idea what you're on about. Your posts aren't making any sense.

[dungeon]

dungeon, I've no idea what you're on about. Your posts aren't making any sense.

It make sense but does not matter, Debian ship with non-free by default anyway so AMD GPUs not working cos of firmwares is not a problem

@AMD If none could initialize firmwares correctly, maybe we should disable amdgpu? But if just one could we should do - let it be?