Hardware encoding failing to start depending on client trying to play (Firefox/Plex web)

Server Version#: 1.23.0.4497
Player Version#: Plex web 4.55.1

Specs are C2750D4i cpu/board. 16GB RAM. Nvidia GTX 1650 SUPER.

So this is a very weird issue as it worked fine and now it doesn’t. I have several clients that I can watch back transcoded files on no issue, these include TV’s/Phones/PC using Edge Browser.

But if I try to watch in Firefox using the same version of plex web as edge, on the same PC no less, playback will never start, except it did as most recently as last night. So curiosly I’ve tested firefox on another PC, it also fails. But a friend manages to have functioning playback in Firefox when he tested. Meanwhile Edge works across two pcs I tested.

Like I said previously this all worked fine and I still playback fine on a Samsung TV/ and Android Phone.

Regardless if a transcode starts or not nvidia-smi dmon will show activity on the GPU and PCIe bus, heres a terminal view of “nvidia-smi dmon -s upt” output vs7vaHS_d.webp (431×482) (imgur.com) where no transocde is happening but then a burst of data over pci and the gpu powerstate increasing but then nothing, here for comparrision is a working transcode with the same output https://i.imgur.com/kvnjcTt.png where the same data over pci and power state increase can be seen followed by encoder and decoder load.

Below are two log cuts of both a failed firefox transcode and a working edge transcode. I really dont know what im supposed to be finding in here to help me narrow this kind of issue down.

edit: I should also this only appears to be a nvenc issue as disabling "Use hardware-accelerated video encoding " allows the gpu to perform the decoding part fine letting the cpu do the encoding.

FAILED_FIREFOX__PlexMediaServer.log (207.5 KB)
WORKING_EDGE__PlexMediaServer.log (316.6 KB)

1 Like

I ran into a similar issue recently on the most recent version of PMS under Docker with a Quadro p2000 gpu. The behavior I noticed is that the “Plex Media Server” instance would start using insane amounts of CPU (600% non-stop which is the container limit), while this same PID would also appear under nvidia-smi as a gpu process. A user also reported to me that “use hardware transcoding” would result in a playback failure.

What’s interesting to me is that the GPU process was never actually a transcoder process, but the server itself.

Regardless, I couldn’t seem to fix the issue so I reverted to the latest stable release (1.22.3) and the behavior seems to have gone away. There were some recent fixes to Nvidia GPUs in the Beta branch, but it looks like work still needs to be done in regards to HW Transcoding.

Are you able to try downgrading to a non-beta version?

1 Like

I attempted downgrading the server to a few previous versions nothing had any affect nor did changing the nvidia driver. I stumbled across this long running thread which has several people with the same issue, hopefully someone either figures out a fix or plex can address it.

I hadn’t noticed any wild CPU usage or rogue processes in nvidia-smi with my testing but maybe ill give it a shot again at the weekend.

This topic was automatically closed 90 days after the last reply. New replies are no longer allowed.