30% persistent Plex CPU load after any user transcodes

[codycrank]

Server Version#: 1.21.1.3830
Player Version#: Any

Plex VM (ESXi), nVidia pass-through, working as expected. Transcodes offload to the GPU and CPU remains nearly idle most of the time. Been running this configuration for ~1 year without issue.

Recently upgraded from Ubuntu 16.04 to 20.04, everything continues to work, but there is a steady 30% cpu load that starts once someone transcodes something and doesn’t go away until Plex is restarted. When the Plex service is initially started, CPU is idle. If I direct stream something, it remains idle. If something transcodes, it offloads to the GPU, CPU spikes to 30% and stays there, even after all streaming is stopped, until the Plex service is restarted.

I’ve done all the database cleanups, refreshed all metadata, disabled my DVR/Tuner. I’ve tried multiple different nVidia drivers. I’ve even rebuilt Plex back to 16.04 thinking the upgrade broke it but the problem persists. It’s as if the transcoder isn’t letting go of the GPU. Has anyone experienced this? I don’t know when it started, I noticed the high CPU one day recently in ESX and started digging in. I do know this behavior is new and nothing changed from a configuration (once I rolled back to 16.04) other than Plex.

[ChuckPa]

without any detailed information (Such as debug log files)…

Audio transcodes can chew up 30 % on a NAS CPU.
Subtitle burning can easily chew up 30% of an Intel i7-class machine.

[codycrank]

This is an old i7-3770, plex given all 4 vCPU’s as the other stuff on the host in low priority. This load persists once all activity is over. It remains days/weeks until plex is restarted. It doesn’t really seem to have much impact, as the GPU is still doing most heavy lifting for video transcodes and most don’t use subtitles. It’s like once plex grabs the GPU, it doesn’t let it go.

Can I PM you the logs? I saw other posts and did as you mentioned stop, start, wait, debug, reproduce, stop download.