2024-05-06, 03:42 PM
(2024-05-06, 10:40 AM)nyanmisaka Wrote: If you don't plan to use iGPU, then don't pass its corresponding renderD128 to docker.
In addition, according to the libdrm code, 129->128 mapping is strongly not recommended, please do 129->129. The cardX is not required.
host:
/dev/dri/renderD128 # iGPU
/dev/dri/renderD129 # dGPU
/dev/dri/card0
/dev/dri/card1
docker:
/dev/dri/renderD129 # dGPU
/dev/dri/card1 # optional, not required
BTW the QSV on Linux is based on VA-API.
Thanks for the info! And yeah, I only pass one device at a time. The situation I described is with only passing my dGPU. I'm aware Linux QSV is Intel's implementation of VA-API. I figured my trying to be smart with 129->128 was probably a pitfall, which is why I changed it. No need to complicate things like that.
Jellyfin 10.10.0 LSIO Docker | Ubuntu 24.04 LTS | i7-13700K | Arc A380 6 GB | 64 GB RAM | 79 TB Storage