Closed comassky closed 5 years ago
Hi!
I can't reproduce it in my environment because I don't use Plex and I haven't Plex Pass.
As it appears to me, OS is irrevelant here (at least it is not distro-specific). Problem somewhere between Plex Server and Nvidia Driver. Power management of Nvidia GPUs is quite strange. Many years ago, when I used to use ffmpeg for video convertations on GPU, I noticed GPU power spikes when processing stops. Sometimes it caused entire system hang because my configuration had weak power supply units.
Observations you made is very useful. They actually prove problem can be workarounded from Plex Server side by means of freeing some GPU resources/handles when no GPU processing active. So workaround is definitely within Plex team grasp. Even if root cause of such power mismanagement is inside Nvidia driver (I believe it is), Plex team can make an effort and expedite this bug to Nvidia support, because they can map a broad picture and describe how to reproduce problem in terms of Nvidia API invocations.
As for nvidia patch project, I don't think we can address this issue in near future. If problem caused by Plex Server, it is out of scope for us. If problem is caused by Nvidia driver - it should be properly implemented in driver code, binary patching is not a substitution for correct implementation. But I'll leave this issue open just in case if someone can propose workaround on how to reset power state without Plex server restart. Maybe nvidia driver utils may help to manage power, I'm not aware about it yet.
Hi Snawoot,
Thanks a lot for your answer.
Actually it was not necessarily a bug for this repository but more a possibility to compare our results to see if the bug was generalized.
I put the bug on the forum plex, which told me that it was an Nvidia bug, but it concerns only plex ...
I hope they will treat tis issue, or pass it on to Nvidia
Just for information, bug seems to be a ffmpeg issue !
More precisely the version ffmpeg used by plex, which an old 3.x.
They are currently working (for the end of the year), on a rise in 4.x which will correct the bug
Thanks! Closing issue.
@comassky we're experiencing something similar in one of our project. Do you give some more information about this ffmpeg issue? I mean some links/issues about this issue in ffmpeg side.
https://ffmpeg.org/pipermail/ffmpeg-devel/2016-June/195598.html
But it's a 2016 issue, bug is fixed since 4.x ...
For the moment, the new Plex transcoder is still ko (https://forums.plex.tv/t/plex-media-server-1-16-7-1597-updated-new-transcoder-preview/451135/137) , even if they upgrade to a recent ffmpeg version.
I see. In our project, it's getting stuck in opening a new session in multithread environment. We use ffmpeg 4.1
Thank you for the information
Okay, interresting !
In Jellyfin project (Plex like), ffmpeg seems to be Ok with this issue, issue is not present.
Maybe a specific management ?
Hi,
I noticed that after running transcodes on plex, with hardware acceleration enabled (gpu), the p-state of my nvidia gtx 1070 gets stuck in P0 mode. Because of this it is drawing a lot of power (more than while transcoding in P2…), and the fan is spinning up unnecessarily as well…
If I stop the plex media server, the p-state immediately gets back to P8 as it should be when idle. Can you please do something with the transcoder to let the gpu return to P8 after having finished the transcoding?
Processes using the GPU while transcoding:
Processes after transcoding has finished:
Somehow Plex Media Server doesn’t let the GPU go…
Plex Teams says it's a Nvidia (430.26) / Kernel (stock 4.19) / OS interaction (Debian Buster) ...
Do you think it's a global bug ? Can you reproduce in your environnement ?