Plex HW Transcoding Fails - QuickSync Un-Responsive with 100% GPU Load

[daves]

Unraid Version: 6.10.2 (same issue with 6.9.x)

CPU: Intel Xeon e2278G

Motherboard: Gigabyte C246-WU4

Dockers Using QuickSync: plexinc/pms-docker, blakeblackshear/frigate:stable-amd64

 

I have run Plex in a docker with hardware transcoding via Intel QuickSync for years without issue.  For the past several months I have the following issue:

1. Plex works fine for several days, hardware transcoding works as expected.
2. Plex fails to play any video file that requires transcoding (direct play continues to work fine)
3. GPU Load is listed at 100% (see below, screenshot taken while frigate and plex docker stopped)
4. Only way I can get it to work again is rebooting server.

 

Any suggestions on how to troubleshoot or recover or fix it?  Thanks!

 

Edited June 6, 2022 by daves

[doobyns]

Downgrade your plex to the last good working version, the v1.24.5.5173, go to your Plex container, then "edit", then set the repository to :

plexinc/pms-docker:1.24.5.5173-8dcc73a59

 

Then click to APPLY, you could have to wait 10 or 15 minutes at the first launch before to access to Plex again, you MUST wait until the end of process, and then you'll have Plex working again with no transcoding issue.

 

I've transcoding issue too, but in my case it's with nvidia hw transcoding...

 

Plex slowly become a piece of crap, coded with the *ss... i'm looking for an alternative which works on smart-tv.
 

[daves]

Hmmmm, this doesn't feel like an issue with Plex.  Transcoding works...until it stops working and a reboot is required.  I suspect that it's an issue with the i915 driver.

Edited June 6, 2022 by daves

[arturovf]

have this issue as well

[esk1]

I just experienced this issue for the first time on 6.10.2.  My symptoms completely matched the OP and a reboot fixed it.  Will monitor....

[daves]

This continues to be a problem for me on 6.10.3.

 

Is there some way to completely re-load the graphics drivers short of rebooting?

Edited August 26, 2022 by daves

[daves]

Continues to be a problem in 6.11.1.