Hi! I’ve virtualized PMS (ubuntu server 18.04.4) over ESX server (NUC) and hardware transcoding was working (doing iGPU passthrough - Intel Iris Graphics 655) for a few months without any problem. Last week I heard NUC fan working at 100% and after a few checks I realized that the plex server was consuming 90-100% of the CPU. I ran intel_gpu_top and it showed that the GPU was apparently being used during transcoding, however CPU usage is still extremely high when previously it was at most between 0-5%.
-
HW transcoding log:
tail -f Plex\ Media\ Server.log | grep -i hardware
Jul 01, 2020 20:56:43.556 [0x7fb988ff9700] DEBUG - Codecs: hardware transcoding: testing API vaapi
Jul 01, 2020 20:56:43.558 [0x7fb988ff9700] DEBUG - Codecs: hardware transcoding: testing API vaapi
Jul 01, 2020 20:56:43.560 [0x7fb988ff9700] DEBUG - Codecs: hardware transcoding: testing API vaapi
Jul 01, 2020 20:56:43.562 [0x7fb988ff9700] DEBUG - Codecs: hardware transcoding: testing API vaapi
Jul 01, 2020 20:56:44.837 [0x7fb989ffb700] DEBUG - TPU: hardware transcoding: using hardware decode accelerator vaapi
Jul 01, 2020 20:56:44.846 [0x7fb989ffb700] DEBUG - TPU: hardware transcoding: zero-copy support present
Jul 01, 2020 20:56:44.846 [0x7fb989ffb700] DEBUG - TPU: hardware transcoding: using zero-copy transcoding
Jul 01, 2020 20:56:44.846 [0x7fb989ffb700] DEBUG - Codecs: hardware transcoding: testing API vaapi
Jul 01, 2020 20:56:44.847 [0x7fb989ffb700] DEBUG - TPU: hardware transcoding: final decoder: vaapi, final encoder: vaapi
Jul 01, 2020 20:56:48.008 [0x7fb989ffb700] DEBUG - TPU: hardware transcoding: using hardware decode accelerator vaapi
Jul 01, 2020 20:56:48.051 [0x7fb989ffb700] DEBUG - TPU: hardware transcoding: zero-copy support present
Jul 01, 2020 20:56:48.051 [0x7fb989ffb700] DEBUG - TPU: hardware transcoding: using zero-copy transcoding
Jul 01, 2020 20:56:48.051 [0x7fb989ffb700] DEBUG - Codecs: hardware transcoding: testing API vaapi
Jul 01, 2020 20:56:48.052 [0x7fb989ffb700] DEBUG - TPU: hardware transcoding: final decoder: vaapi, final encoder: vaapi -
intel_gpu_top output:
render busy: 43%: ████████▋ render space: 17/4096 task percent busy GAM: 80%: ████████████████ vert fetch: 0 (0/sec) CS: 43%: ████████▋ prim fetch: 0 (0/sec) TSG: 38%: ███████▋ VS invocations: 0 (0/sec) VFE: 17%: ███▌ GS invocations: 0 (0/sec) GAFS: 9%: █▉ GS prims: 0 (0/sec) RS: 0%: CL invocations: 0 (0/sec) TDG: 0%: CL prims: 0 (0/sec) SVG: 0%: PS invocations: 0 (0/sec) CL: 0%: PS depth pass: 0 (0/sec) SF: 0%: VF: 0%: GAFM: 0%: VS: 0%: SOL: 0%:
I tried a lot of things to fix the issue:
- Disable and enable iGPU from ESX and rebind to PMS virtual machine
- Reapply advanced virtual machine params:
- svga.present: FALSE
- pciHole.start: 2048 (virtual machine has 4GB RAM)
- Install a new PMS over Debian 10
- Updating ESX to 7.0 (clean install)
Anything was helpful, CPU load is high and I’m feel a little bit frustrated.
Can anyone help me?
Thanks!