Agreed. I have to “systemctl restart plexmediaserver” at the end of each day to reset the GPU back to P8. Any time a transcode happens, it gets stuck in P0 mode. It’s a bit annoying to tell the truth.
Is there any updates on this issue @chrisallen ?
Nothing yet sorry.
1 Like
Really hope this annoying bug can be fixed soon. It’s not just power hungry (~27W extra in idle), my GPU (Quadro P4000) gets 12C hotter with P0 state.
Mine is also stuck at P0 100% of the time 
@chrisallen is there any news on this topic? Seems to me that this bug is very low in prio list.
There are a number of other larger issues being worked through at this time. We are still aware of this issue and it’s impact to users.
Can confirm the issue still persists on Version 1.18.2.2058. Just tried today Plex on Unraid with a P1000 and my GPU was stuck.
Thanks for keeping us informed you guys. Hopefully with all the new toys going around at the end of the year, we can get some interest on GPUs again.
Also getting this issue with a GTX 1080
Causes the GPU to overheat and stop working.
lsof /dev/nvidia* | awk ‘{print $2}’ | xargs -I {} kill {} | nvidia-smi -pm 1
Fixes the issue temporarily - but will kill any active gpu transcodes. So I can only use it manually when the GPU has crashed.
Very frustrating…
Whilst this script kind of works and you could certainly run it from root’s crontab to spawn a few times a day to automate putting the GPU back into P8 state (it DOES leave active Plex transcodes and other non-Plex GPU processes alone, and it DOES successfully kill the Plex process that has not released the GPU) unfortunately the process that it kills also happens to be the main parent Plex server process meaning that you need to restart Plex cleanly anyway, otherwise (in my case at least) the Plex server is rendered unavailable after this script has successfully run unless I manually intervene…
I’ve added in ;service plexmediaserver restart to line 25 at the end (after kill -9 "$TOKILL") which seems to do the trick on my Ubuntu 19.10 server, but it does take precisely 90 seconds for it to all come alive again, as I think systemd is potentially waiting on something before restarting everything cleanly. To get around this, you can edit /etc/systemd/system.conf and reduce the DefaultTimeoutStopSec to say 10s, and run sudo systemctl daemon-reexec for the settings to take effect.
I don’t know how to manually intervene to tell the Plex process to “let go” of the GPU otherwise without killing it…I suspect it would need to be done in the code of the Plex process itself. If anyone knows an alternative workaround it would be good.
Is there any more news on a fix for this from the Plex devs? It certainly seems to happen to a wide range of Nvidia GPUs on Linux (all types?). My GTX 1650 does this and it’s a bit annoying when you’re burning watts unnecessarily when idle…
Hi, Is this annoying bug being worked on yet?
Hi guys, Unfortunately, I don’t yet have anything new to report on this issue.
I have 2 GPU’s in my server, the first one is a gt710 for the Unraid GUI boot and its most of the time on P8.
the second one, a GTX1050TI is for Plex transcode and its stays at P0 while its doing nothing.
command “fuser -kv /dev/nvidia*” works fine, after using it it switched to P8 (its else resetting the gt710 what is not needed…) also i can start a new transcoding stream without problems… (its switchs back to P0)
Plex media server is on version 1.18.6.2368
and am using the Unraid Nvidia plugin (currently on 6.8.2)
this is for what i can found out a Plex server problem, plex is doing stuff whats used the gpu, after using that fuser command its kills the plex processes… and those plex processes use 10MB Vram…
1 Like
+1 I have the same problem. Using RTX 2070 and Unraid 6.8.2 with Plex 1.18.6.2368 in docker. Video card is always in p0 state consuming 42w. Using “fuser -kv /dev/nvidia*” command switches the card into p8 state with 16w consumption. Once I start Plex (even without transcoding) it switches to p0 state back and never comes back to p8. As I read it is somehow related to ffmpeg version that Plex uses. Any plans to fix it? As additional 24/7 25-30w consumption is costly here in Germany.
+1 the same problem here. I just bought extra vga and running free month of FlexPass after which I will aim to buy lifetime subscription. But, without this working correctly, this is not gonna happen.
Dear Plex, any news on this problem or at least acknowledgement and working on the issue?
We have not forgotten this issue, and are working with Nvidia to understand this issue, and why it the power state doesn’t return to a lower value when the Plex Media Server process closes it’s request to the GPU.
Thanks for quick reply! Hope you find a solution soon! Ready to assist with testing if needed.
Here is a script with the nvidia-smi querys to check the pstate and if a process is running so you can cron the script every few minutes to kill if there is a bad process running without the problem of killing a current transcode.
#!/bin/bash
gpupstate=$(nvidia-smi --query-gpu="pstate" --format=csv,noheader);
gpupid=$(nvidia-smi --query-compute-apps="pid" --format=csv,noheader);
if [ "$gpupstate" == "P0" ] && [ -z "$gpupid" ]; then fuser -kv /dev/nvidia*; fi;
if you have more than one nvidia gpu you have to add “-i 0” (where 0 is the id number of the gpu that transcodes plex) at the end of the line on the querys.
4 Likes
