GPU Passthrough for Unraid
Prerequisites
- NVIDIA GPU (GTX 10-series or newer recommended)
- Unraid 6.12+
Step 1: Install Nvidia Driver Plugin
- Go to Apps → Search "Nvidia Driver"
- Install Nvidia-Driver by ich777
- Reboot Unraid
Step 2: Verify GPU Detection
Open Unraid terminal and run:
You should see your GPU listed:
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 550.xx Driver Version: 550.xx CUDA Version: 12.x |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
|===============================+======================+======================|
| 0 NVIDIA GeForce ... Off | 00000000:XX:00.0 Off | N/A |
+-------------------------------+----------------------+----------------------+
Step 3: Container Configuration
The Stash Sense template automatically includes GPU settings.
If configuring manually, ensure these are set:
| Setting | Value |
|---|---|
| Extra Parameters | --runtime=nvidia --gpus all |
| NVIDIA_VISIBLE_DEVICES | all (or specific GPU UUID) |
Troubleshooting
"nvidia-smi: command not found"
Driver plugin not installed or Unraid needs reboot.
# Check if plugin is installed
ls /boot/config/plugins/nvidia-driver.plg
# If missing, reinstall from Apps
"No GPU detected in container"
Check container has GPU access:
If this fails, verify:
- Extra Parameters includes
--runtime=nvidia - Container was recreated after driver install (not just restarted)
"CUDA out of memory"
Another process is using the GPU. Check what's using it:
Look at "Processes" section. Common culprits: - Plex hardware transcoding - Tdarr - Frigate - Other ML containers
"Failed to initialize NVML"
The Nvidia container toolkit isn't configured. Re-install the driver plugin and reboot.
Selecting a Specific GPU (Multi-GPU Systems)
If you have multiple GPUs, specify which one to use:
# List GPU UUIDs
nvidia-smi -L
# Use specific GPU
NVIDIA_VISIBLE_DEVICES=GPU-xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx
Or by index: