Transcoding broken after latest update (only on certain files)

Server Version#: 1.30.1.6562
Player Version#: 4.98.2

UnRAID server with PMS running in a docker using an nVidia 1660 Ti with driver 525.78.01.

After the auto update of the Plex container from 1.30.0.6486 to 1.30.1.6562, nVidia-SMI shows a transcode when trying to play the file but the player only sits on the loading screen with a spinning cursor and never plays it (but the Dashboard shows it as playing).

If I uncheck “Use hardware-accelerated video encoding” then file plays fine (but obviously doesn’t use the GPU for encoding).

Log snippet:
Jan 11, 2023 10:21:34.936 [0x14cee1a38b38] Error — [Req#1998c/Transcode/mff58ubc8ggvwc53purztpeh/9f425c66-15e7-42f6-83d0-3f444db6dae1] Error while decoding stream #0:0: Generic error in an external library
Jan 11, 2023 10:21:34.937 [0x14ceefeecb38] Error — [Req#1998f/Transcode/mff58ubc8ggvwc53purztpeh/9f425c66-15e7-42f6-83d0-3f444db6dae1] [hevc @ 0x146624574280] Could not find ref with POC 14
Jan 11, 2023 10:21:34.938 [0x14ceee87db38] Error — [Req#19992/Transcode/mff58ubc8ggvwc53purztpeh/9f425c66-15e7-42f6-83d0-3f444db6dae1] [hevc @ 0x146624574280] No decoder surfaces left
Jan 11, 2023 10:21:34.939 [0x14cee1a38b38] Error — [Req#19995/Transcode/mff58ubc8ggvwc53purztpeh/9f425c66-15e7-42f6-83d0-3f444db6dae1] [hevc @ 0x146624574280] decoder->cvdl->cuvidDecodePicture(decoder->decoder, &ctx->pic_params) failed → CUDA_ERROR_INVALID_VALUE: invalid argument
Jan 11, 2023 10:21:34.939 [0x14ceefeecb38] Error — [Req#19998/Transcode/mff58ubc8ggvwc53purztpeh/9f425c66-15e7-42f6-83d0-3f444db6dae1] [hevc @ 0x146624574280] hardware accelerator failed to decode picture

Plex Media Server Logs_2023-01-11_14-17-12.zip (2.5 MB)

3 Likes

Same problem here…

annyoing
Nvidia RTX A2000 here…

nvidia driver version: 525.78.01
debian buster

Same problem, same error messages

NVIDIA GeForce GTX 1050 Ti
Nvidia Driver Version: 525.78.01

Same exact problem here with the same log output.

NVIDIA GeForce 1060
Driver 525.78.01 - tried rolling this back but did not help so came back to current

Hi,

Same problem here
Running UnRAID
Nvidia RTX A5000
Plex on docker container, some HEVC files (not all) not playing at all, stuck buffering at the beginning of the files but never starts.

This happens since version 1.30.1.6562 (linuxserver/plex:1.30.1)
Rolled back to 1.30.0.6486 (linuxserver/plex:1.30.0), everything works fine with this version.

1 Like

The Nvidia 525+ drivers cause problems with PMS on Linux. Might try rolling back to 515.86.01.

I saw that over in the other thread. I did the roll back on drivers to 515.86.01 but it still did not work with any 1.30.x version of PMS. Reverting back to 1.29 solves the issue for now.

Was the host restarted after downgrading the Nvidia driver ?
(The kernel won’t unload 525 once loaded)

Correct. Host was rebooted. I also rebooted between each attempt of the 1.30.x versions. 1.29.0 is stable with the 515.86.1 drivers.

@mwieger

A couple things. What type media are you trying to play? HEVC and H.264?

I’m able to confirm HEVC, HDR, with anything 515.86.01 and the version before on 1.30.2

I tested this on Ubuntu with Plex docker and LSIO docker, QNAP, and Ubuntu Server native.

All of my media is HEVC and under 1.30.x with rolled back NVIDIA drivers I still get the same error output as the original poster showed. 1.29 fixes the issue.

I will say it seems to be triggered at the beginning of play when the system is determining to transcode or not. Manually setting playback to 2mbps transcode does not trigger the issue. I say that because I do not think it is a codec problem, because the system will transcode if manually told to do so. So something between 1.29 and 1.30.x changed something in this sequence.

Hi All,
I originally follow the video by SpaceInvaderOne on YouTube for setting up my Nvidia Hardware Decoding on Unraid for PMS, it has been working for years, but not it doesn’t work after a recent upgrade to Plex or Unraid. I have “Use Hardware Acceleration when available” checked under Transcoder and I have entered my GPU ID into the Plex Docker Container settings along with --runtine=nvidia. I don’t really know what else is required. I wish I could find a simple 10 step guide to enabling this it seems so confusing having to use 3rd party sources to figure this out. I need some help, this shouldn’t be this hard to get HW Transcoding to work. Nvidia Driver Version: 515.86.01

They are having an issue with it on the latest versions. We have to roll back to pms 1.29 for now.

Thanks, but I have no idea how to roll back PMS on Unraid. I might have to wait for PMS 1.31… as I only know how to roll up.

Not sure which repository you are using but you do it by hitting edit on the docker and adding a colon and the version behind it. Like I did here.

1 Like

To all following here:

  1. We have identified the issue with Nvidia 525+ drivers and now have a candidate change to PMS.

  2. We identified the issue with certain 1080p input files failing to HW transcode.

I have both DEB and RPM files (Engineering development – No QA) for testing.

There will be another update before we complete this work

I would appreciate feedback at your earliest convenience.

Thanks.

Just tried the latest Public version (1.31.1.6782) and still can’t play certain files, I’m getting the “No decoder surfaces left” error and “out of memory” errors. nVidia driver v525.78.01 I rolled back to 1.29.2.6364 and issues no longer occur.

Need logs ZIP file which captures this failure

If possible, a sample of the file so I may recreate it

To roll back or forward in docker,

  1. With /config mapped to persistent storage (not deleted)
  2. Delete the container
  3. Change the spec to call for a specific image version, e.g.
  4. Create the container again ON TOP OF the original container.

(example from a docker run)

-v /vie:/data \
-v /glock:/glock \
--device=/dev/dri:/dev/dri \
plexinc/pms-docker:1.29.2.6364-6d72b0cf6 

Will get those files for you. I’ll let you know when I have them and will send a PM with a GDrive link to the file. Thank you!