Hardware acceleration

[technically_artist]

Hi,

I have recently gotten started on docker, jellyfin and self-hosting, so apologies in advance if I miss some things.

As I mentioned in the description, I am running Jellyfin through CasaOS.
However, even after I have enabled hardware acceleration through the Jellyfin admin dashboard (Intel Quicksync QSV) I noticed that CPU usage is still through the roof at nearly 90%.
My assumption was that with hardware acceleration transcoding would be delegated to the integrated GPU (a Kabylake 630) and not tax the CPU itself too harshly.

I feel like I missed something during setup. I did add the "/dev/dri" bit to the container, but that doesn't seem to make a difference.

Are there any obvious (or less obvious) things I might've missed?

[TheDreadPirate]

Can you share your jellyfin and ffmpeg logs? We need to verify that ffmpeg successfully used the hardware. I'm leaning towards "probably". Since you enabled QSV and it didn't fail to play, it is probably transcoding with QSV. High CPU usage isn't unusual. Keep in mind that even with hardware accelerated transcoding, your CPU still has to feed and orchestrate the GPU's activities.

Again, the jellyfin and ffmpeg logs will guide us.

[technically_artist]

Can you share your jellyfin and ffmpeg logs?  We need to verify that ffmpeg successfully used the hardware.  I'm leaning towards "probably".  Since you enabled QSV and it didn't fail to play, it is probably transcoding with QSV.  High CPU usage isn't unusual.  Keep in mind that even with hardware accelerated transcoding, your CPU still has to feed and orchestrate the GPU's activities.

Again, the jellyfin and ffmpeg logs will guide us.

These are the ffmpeg and jellyfin logs, at a quick glance I couldn't find anything about hardware encoding yet.

[TheDreadPirate]

From your ffmpeg log.

Stream #0:2 -> #0:0 (hevc (native) -> h264 (h264_qsv))

Definitely hardware transcoding. Again, high CPU usage is not unusual even with hardware accelerated transcoding. Especially on older CPUs.

[bitmap]

You can also double check using intel_gpu_top, which is part of the intel-gpu-tools package. Start playback of a video that requires transcoding and on your server, run intel_gpu_top (may require sudo) to see any activity on your iGPU. Not useful for much else, but can certainly help in verifying for absolute certain that things are working as expected.