Since I updated to 21.08 - PLEX no GPU transcoding

[Janus0006]

Hi,

Since I updated to 21.08, My Plex seems to be unable to use my NVIDIA P600 to transcode.

From my host, nvidia-smi is not working
NVIDIA-SMI has failed because it couldn't communicate with the NVIDIA driver. Make sure that the latest NVIDIA driver is installed and running.

"lspci" give me: 02:00.0 VGA compatible controller: NVIDIA Corporation GP107GL [Quadro P600] (rev a1)

From my docker apps, I can Allocate 1 nvidia/gpu GPU (at least, he know a nvidia gpu is available)

In Plex console, I can see the errors:
[0x7f511d084b38] [Transcode] [FFMPEG] - Cannot libcuda.so.1
[0x7f511d084b38] [Transcode] [FFMPEG] - Could not get dynamically load CUDA

Seems to do not find the NVIDIA/Cuda drivers, but I tried to reinstall them with the same issue
apt install nvidia-cuda-dev nvidia-cuda-toolkit

And, as my app can see the NVIDIA GPU to allocate, it does not means the card/driver are seen ?

Before the update, everything was working fine.

Does anyone had this kind of issue after upgrading ? Or have you an idea on how to solve this ?

Many thanks

 

[Kris Moore]

Locally my 21.08 version still seems to be transcoding properly. Can you please open a ticket on https://jira.ixsystems.com and attach a debug file? (System -> Advanced -> Download Debug). We'll investigate and see if a fix is possible. I'm wondering if that Quadro P600 isn't attaching to the nvidia driver properly.

 

[michael shaw]

I got same problem. also NVIDIA P600
nvidia-smi
NVIDIA-SMI has failed because it couldn't communicate with the NVIDIA driver. Make sure that the latest NVIDIA driver is installed and running.

And in vm(centos7) after nvidia driver installed

lspci | grep -i vga
00:02.0 VGA compatible controller: Red Hat, Inc. QXL paravirtual graphic card (rev 05)
00:06.0 VGA compatible controller: NVIDIA Corporation GP107GL [Quadro P600] (rev a1)
and running nvidia-smi reply:
Unable to determine the device handle for GPU 0000:00:06.0: Unknown Error

anyone can help?

 

[Trexx]

I got same problem. also NVIDIA P600
nvidia-smi
NVIDIA-SMI has failed because it couldn't communicate with the NVIDIA driver. Make sure that the latest NVIDIA driver is installed and running.

And in vm(centos7) after nvidia driver installed

lspci | grep -i vga
00:02.0 VGA compatible controller: Red Hat, Inc. QXL paravirtual graphic card (rev 05)
00:06.0 VGA compatible controller: NVIDIA Corporation GP107GL [Quadro P600] (rev a1)
and running nvidia-smi reply:
Unable to determine the device handle for GPU 0000:00:06.0: Unknown Error

anyone can help?

So you are running a VM, which passing in a GPU, which is then trying to pass the Passthrough GPU into a Kubernetes Cluster to a containerized version of Plex...???

 

[Janus0006]

Locally my 21.08 version still seems to be transcoding properly. Can you please open a ticket on https://jira.ixsystems.com and attach a debug file? (System -> Advanced -> Download Debug). We'll investigate and see if a fix is possible. I'm wondering if that Quadro P600 isn't attaching to the nvidia driver properly.

Many thanks for your answer. JIRA is opened "NAS-112466"

 

[darkmode]

I wonder if it's the cgroups v2 issue while using the nvidia container runtime.

cgroup issue with nvidia container runtime on Debian testing · Issue #1447 · NVIDIA/nvidia-docker

1. Issue or feature description Whenever I try to build or run an NVidia container, Docker fails with the error message: docker: Error response from daemon: OCI runtime create failed: container_lin...

github.com

Manual setup of nvidia-docker2 for Debian 11? · Issue #1537 · NVIDIA/nvidia-docker

Is it possible to setup nvidia-docker2 before any stable release? Or are there any plans in the near future to bring support to Debian 11?

github.com

 

[Janus0006]

I have a bit of difficulties to understand the nvidia-docker ...
On previous versions, the drivers was installed on the Truenas OS and passed trough to the docker, Am I right ?
Now you seems to say the driver is "running" via a container ?

Recently, I tried to reinstall nvidia/cuda drivers in the OS and the "apt update" bugged all the OS, unable to boot at all, in recovery mode. I had to reinstall TS 21.08 and reimport config.

Is there a complete guide, somewhere on "How to configure a GPU to be used in a docker apps" ? Depending on the version you have. I searched but I did not found anything

 

[Zandr Milewski]

I'll chime in with a possibly related issue. Since upgrading to 21.08-BETA.1, I haven't been able to assign a GPU to an App container. The GPU Configuration header is present when installing or editing a container, but there are no fields below it.

 

[Janus0006]

FYI, I was able to fix my issue.
The issue appears to be related to the Plex app from the Official catalog. As soon as I configured the apps with the Truechart catalog version, everything works fine.