Plex Hardware Transcoding on Nixos (Docker Compose)

Plex Hardware Transcoding on NixOS (Docker)

The Problem

Plex running in Docker on a NixOS host could detect the Nvidia GPU (via nvidia-smi) but failed to use it for transcoding. The dashboard showed software transcoding (high CPU usage), and logs reported [TPU] hardware transcoding: enabled, but no hardware decode accelerator found.

The Root Causes

1. Driver Injection (CDI): NixOS uses the newer CDI (Container Device Interface) method, requiring specific deploy blocks rather than the legacy gpus: all flag.
2. Permissions: The container user (PUID=1000) lacked permission to access the Direct Rendering Manager (DRM) nodes (/dev/dri/card0 and /dev/dri/renderD128).
3. Library Paths: The Nvidia driver libraries (e.g., libnvcuvid.so) were injected into non-standard paths on NixOS (/usr/local/nvidia/lib64), which Plex does not check by default.

---

The Solution

1. NixOS Host Configuration (configuration.nix)

Ensure the host has graphics enabled and the Nvidia container toolkit active.

hardware.graphics.enable = true;
hardware.nvidia.modesetting.enable = true;
hardware.nvidia-container-toolkit = {
  enable = true;
  mount-nvidia-executables = true;
};

**2. Final Working docker-compose.yml**

This configuration addresses the driver injection, permission groups, and library paths.

services:
  plex:
    image: lscr.io/linuxserver/plex:latest
    container_name: plex
    network_mode: host
    environment:
      - PUID=1000
      - PGID=1000
      - TZ=Etc/UTC
      - VERSION=docker
      
      # 1. Enable Driver Capabilities
      - NVIDIA_VISIBLE_DEVICES=all
      - NVIDIA_DRIVER_CAPABILITIES=all
      
      # 2. Force Plex to check NixOS/CDI library paths
      - LD_LIBRARY_PATH=/usr/lib/plexmediaserver/lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/lib:/usr/lib64
      
    # 3. Grant access to host hardware groups
    # Run `getent group render` and `getent group video` on host to confirm IDs.
    group_add:
      - "303"  # render group (for /dev/dri/renderD128)
      - "26"   # video group (for /dev/dri/card0)
      
    volumes:
      - ./Library:/config/Library
      # 4. Map the hardware device nodes
      - /dev/dri:/dev/dri
      # ... (media volumes)
      
    restart: unless-stopped
    
    # 5. Use CDI for GPU resource allocation
    deploy:
      resources:
        reservations:
          devices:
            - driver: cdi
              device_ids:
                - nvidia.com/gpu=0
              capabilities: [gpu]

Troubleshooting Commands

If the GPU fails again, use these commands to diagnose where the chain is broken:

• Check GPU Visibility:

docker exec -it plex nvidia-smi

• Check Permissions:

docker exec -it plex ls -ln /dev/dri
# Ensure the group IDs match what you set in group_add

• Find Missing Libraries:

docker exec -it plex find / -name "libnvcuvid.so*"
# If found in /usr/local/nvidia/lib64, ensure LD_LIBRARY_PATH includes it.