[Nouveau] Any tips to avoid Nouveau crashes & freezes?

[Nouveau] Any tips to avoid Nouveau crashes & freezes?

[Ian Kelling]

[-- Attachment #1: Type: text/plain, Size: 1309 bytes --]

At FSF, we've been having trouble with Nouveau related kernel errors and
freezes which require a reboot for about 2 years now.

The issue mostly occurs when doing video conferencing in a browser plus
recording that using OBS (Open Broadcast Studio). We've found all the
configurations we've tested to crash/freeze from about once a day to
several times a day. This is mainly a problem when we are live streaming
our annual LibrePlanet conference, which is happening next weekend.

More info:

Our kernels don't load any nonfree firmware blobs. Currently they run
Trisquel 11.

We've tried various Nvidia cards, about 3 different generations.

For video conferencing, we use webrtc through BigBlueButton (versions
with older free MongoDB) on chromium (it's webrtc video has been more
performant than mozilla based browsers).

I've attached a kernel log of an example of a freeze. At the end of the
log we could ssh in, but the display was frozen.

Any tips to help solve this are welcome. Eg: what model or generation of
cards you think is more stable, or any software versions or
configurations that are more likely to be stable, or any general
ideas. 

Related note: Trisquel 11 just added support for running a lot of older
amd GPUs without loading nonfree firmware, so we are testing their
stability now.


[-- Attachment #2: GK107-freeze.log.bz2 --]
[-- Type: application/octet-stream, Size: 26484 bytes --]

[-- Attachment #3: Type: text/plain, Size: 169 bytes --]


-- 
Ian Kelling | Senior Systems Administrator, Free Software Foundation
GPG Key: B125 F60B 7B28 7FF6 A2B7  DF8F 170A F0E2 9542 95DF
https://fsf.org | https://gnu.org

Re: [Nouveau] Any tips to avoid Nouveau crashes & freezes?

[Ian Kelling]

Small update: Using llvmpipe for running OBS seems to fix the crashes &
hangs. It uses a lot of CPU, like 7.5 cores vs around 1 without, but
that seems fine for our purpose. Running OBS AMD gpus uses this much cpu
without explicitly asking for llvmpipe.