[FFMPEG] - Cannot load libcuda.so.1 | Hardware encoding in docker on a proxmox vm

Plex Media Server
,
1

Server Version#: 1.43.0.10492

This is my setup

  1. Computer with GTX 1060
  2. Installed on it is Proxmox
  3. Proxmox is running a VM (PCIe passthrough seemingly works no errors other than plex)
  4. Plex is running in docker on that VM

Here is my plex log file

Plex Media Server Logs_2026-02-26_20-02-45.zip (1.3 MB)

I saw [FFMPEG] - Cannot load libcuda.so.1 in the log and was able to find this guy on reddit in much the same situation as me, but i was unable to find an answer from that thread.

Here is my compose.yml:

image
image406×631 20.7 KB

I have the latest drivers installed with sudo ubuntu-drivers autoinstall
nvidia-smi returns this:

+---------------------------------------------------------------------------------------+
| NVIDIA-SMI 535.288.01             Driver Version: 535.288.01   CUDA Version: 12.2     |
|-----------------------------------------+----------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |         Memory-Usage | GPU-Util  Compute M. |
|                                         |                      |               MIG M. |
|=========================================+======================+======================|
|   0  NVIDIA GeForce GTX 1060 6GB    Off | 00000000:00:10.0 Off |                  N/A |
|  0%   31C    P8               5W / 156W |      2MiB /  6144MiB |      0%      Default |
|                                         |                      |                  N/A |
+-----------------------------------------+----------------------+----------------------+
                                                                                         
+---------------------------------------------------------------------------------------+
| Processes:                                                                            |
|  GPU   GI   CI        PID   Type   Process name                            GPU Memory |
|        ID   ID                                                             Usage      |
|=======================================================================================|
|  No running processes found                                                           |
+---------------------------------------------------------------------------------------+

I know this might not be the right place to ask but i thought somebody might have experience with this exact setup.

2

Did you ever resolve this? I’m also getting this error and have made absolutely zero headway towards fixing it.

ERROR - [Req#397/Transcode] [FFMPEG] - Cannot load libcuda.so.1
ERROR - [Req#397/Transcode] [FFMPEG] - Could not dynamically load CUDA

3

Yes i did. My issue was that i did not fully understand all the steps needed to get an nvidia gpu to work inside a docker container

Just to repeat my setup:

  1. Proxmox on machine with nvidia GPU
  2. Ubuntu VM
  3. Docker running plex on that VM.

I first thought i had done something wrong in the passthrough of the device to the VM with IOMMU and driver blacklisting and so on, but running nvidia-smi worked.
My issue was with docker.

This is my new compose file: (some sections may be obsolete, but the moment it started working i stopped changing it)

services:
  plex:
    image: lscr.io/linuxserver/plex:latest
    container_name: plex
    restart: unless-stopped
    networks:
      - default
    runtime: nvidia
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu,video]
    ports:
      - "32400:32400/tcp"
      - "3005:3005/tcp"
      - "8324:8324/tcp"
      - "32469:32469/tcp"
      - "1900:1900/udp"
      - "32410:32410/udp"
      - "32412:32412/udp"
      - "32413:32413/udp"
      - "32414:32414/udp"
    volumes:
      - ./config:/config
      - /data:/data
      - ./transcode:/transcode
    environment:
      TZ: "My_TZ"
      PLEX_CLAIM: "My_Key"
      PLEX_UID: "1000"
      PLEX_GID: "1000"
      NVIDIA_VISIBLE_DEVICES: all
      NVIDIA_DRIVER_CAPABILITIES: compute,video,utility

I kept getting errors when i included this section, so i just didn’t include it originally.

    runtime: nvidia
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu,video]

Turns out that section is crucial, you just need the NVIDIA Container Toolkit for it to work.
I followed this official guide from NVIDIA.

This may not be the solution to your issue, so if you have further trouble go ahead and ask.

4

Thanks for replying! My issue turned out to be something different. I was able to resolve it using the workaround highlighted in this GitHub issue.