Having Vega 10 Linux GPU Hangs? Try Rolling Back The Firmware

[M@yeulC]

Mmm, I didn't really have any performance problems that I noticed (except the long loading times). The hang happened every time I played the game (thus ending my play session), anywhere from 15min to 4hours trough.

debianxfce, well, windows game or not, the GPU shoudln't hang. The kernel should be able to recover (and I think I enabled amdgpu recovery), and the compiler shouldn't emit code that makes it hang in the first place. My question was more: "what is the appropriate bug tracker".

[bridgman]

My question was more: "what is the appropriate bug tracker".

https://www.x.org/wiki/RadeonFeature/#index12h2

Mesa 3D driver (SI-VI) - Mesa : Drivers/Gallium/radeonsi

Probably Mesa3D under radeonsi is the best place to start if the problem is with a specific game.

BTW I don't think debianxfce was saying that hanging is OK because it's a Windows game, just that being a Windows game means that there are other areas worth looking into when troubleshooting like wine settings.

[shmerl]

BTW I don't think debianxfce was saying that hanging is OK because it's a Windows game

Is Vega GPU reset supposed to work in newer kernels? I.e. should it prevent such kind of hangs, or they still can happen?

[TheVE]

I have had random GPU hangs on my RX480 since I got it. Don't hold your breath on this.

[thxcv]

Just tried the December 13th build of Suse Tumbleweed last weekend with a Polaris 10 card. The distro was still on Mesa 18.1.7 by default. So far so good. But once I installed Mesa 18.3.1 from the Suse Factory repo, the booting process stopped with a hard lockup where the GUI was supposed to start. Ah, Linux makes sometimes such a unfinished and untested impression... but Windows 10 Insider Previews (and even stable releases) are no longer that much better though.

So you are saying that you used a rolling release plus changed substantial stuff (mesa) that was actually tested and deemed not ready for primetime but then are wondering that it has been a rough ride?
The opensuse devs held back Mesa 18.2 and 18.3 for good reason but fortunately it was finally released today.

[ms178]

So you are saying that you used a rolling release plus changed substantial stuff (mesa) that was actually tested and deemed not ready for primetime but then are wondering that it has been a rough ride?
The opensuse devs held back Mesa 18.2 and 18.3 for good reason but fortunately it was finally released today.

Well, the only reason which I saw in the discussions there was a GCC bug which prevented shaders to be compiled as they should and this was fixed with a newer GCC release a couple of weeks ago. And the Suse site also mentioned that the particular package which I used was an official package albeit not shipping it through their package management tool yet. But to answer your rhetorical question more broadly: Yes, I expect a major release to work out-of-the-box after it was stamped "official". That might be not the standard in the Linux world and I know that software development is tough, but as an end-user who wants to live on the bleeding edge while also enjoing a reasonable amount of stability, this could be achieved with a little more testing effort to answer the basic question: Does it light up or not?

[thxcv]

Well, the only reason which I saw in the discussions there was a GCC bug which prevented shaders to be compiled as they should and this was fixed with a newer GCC release a couple of weeks ago. And the Suse site also mentioned that the particular package which I used was an official package albeit not shipping it through their package management tool yet. But to answer your rhetorical question more broadly: Yes, I expect a major release to work out-of-the-box after it was stamped "official". That might be not the standard in the Linux world and I know that software development is tough, but as an end-user who wants to live on the bleeding edge while also enjoing a reasonable amount of stability, this could be achieved with a little more testing effort to answer the basic question: Does it light up or not?

Well, as you said it is bleeding edge and also mesa 18.3.1 was still tagged 'experimental' until very recently and not at all out-of-box but rather test it if you want but thats your problem then (as you said yourself it was still shipped with 18.1.7)... so i think the conclusion that it is unfinished and untested does not fit the scenario at all...
There was the GCC problem that you mentioned but i think also some applications that broke with mesa 18.2 and after that also a building problem with LLVM7.

[ms178]

Well, as you said it is bleeding edge and also mesa 18.3.1 was still tagged 'experimental' until very recently and not at all out-of-box but rather test it if you want but thats your problem then (as you said yourself it was still shipped with 18.1.7)... so i think the conclusion that it is unfinished and untested does not fit the scenario at all...
There was the GCC problem that you mentioned but i think also some applications that broke with mesa 18.2 and after that also a building problem with LLVM7.

I understand your position but I am quite sure that it wasn't flagged experimental anymore on December 14th, if it were, I would not complain at all.

As I am not reliant on a stable system currently and wanted to experiment with Linux as a daily driver, I am still a bit grumpy that there are still so many rough edges which prevent me to leave Windows fully behind - even for office use. The Chromium VAAPI related problems with AMD which I mentioned in another thread today, also fall into this category. Great to have this enabled by default now by Suse, but if AMD GPUs are known to have problems then blacklisting them until these issues are solved would have been the wiser choice. There are so many other little things accumulating which could be solved by more sane decisions of the developers or distros in the first place.