Skip to main content

Other Info

NVIDIA Setup on Linux (Debian) Based Systems with a docker container​

I followed this page for the latest NVIDIA Container Toolkit instructions and installing. nvidia-ctk is a utility built to give nvidia graphics capabilities into containers seamlessly.

Uninstalling nvidia drivers from your system can be done like this.

A quick install of the nvidia drivers for Ubuntu 22.04 can be found here

nvidia-ctk should setup the /etc/docker/daemon.json or the ~/.config/docker/daemon.json for you, but this is what my working config looks like for reference:

{
  "default-runtime": "nvidia",
  "features": {
    "cdi": true
  },
  "runtimes": {
    "nvidia": {
      "args": [],
      "path": "nvidia-container-runtime"
    }
  }
}

Be sure to set the devices in your docker-compose.yml to be like this for the emby container to recognize them in the nvidia-smi

devices:
  - /dev/nvidia-uvm:/dev/nvidia-uvm # Added nvidia devices here
  - /dev/nvidia-uvm-tools:/dev/nvidia-uvm-tools # Added nvidia devices here
  - /dev/nvidia-modeset:/dev/nvidia-modeset # Added nvidia devices here
  - /dev/nvidiactl:/dev/nvidiactl # Added nvidia devices here
  - /dev/nvidia0:/dev/nvidia0 # Added nvidia devices here
  - /dev/nvidia1:/dev/nvidia1 # Added nvidia devices here (i have a 2nd GPU so i needed this too)
  - /dev/dri:/dev/dri # For VAAPI access, not necessarily for NVIDIA cards

NVIDIA Quick Reference​

If you ever lose your nvidia transcoding from emby, you can try a few things:

  • Reboot the system. This will solve it 99% of the time
  • Reinstall the NVIDIA Drivers from the web
  • Reinstall the NVIDIA Drivers from the apt repository
sudo apt-get install nvidia-driver-535 -y
# or install nvidia-driver-### where your number is the current version
  • Run nvidia-smi (Output should be similar to below)
$ nvidia-smi
Fri Apr  5 14:58:16 2024
+---------------------------------------------------------------------------------------+
| NVIDIA-SMI 535.161.07             Driver Version: 535.161.07   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 1080        Off | 00000000:24:00.0 Off |                  N/A |
|  0%   41C    P8              11W / 180W |      2MiB /  8192MiB |      0%      Default |
|                                         |                      |                  N/A |
+-----------------------------------------+----------------------+----------------------+
|   1  NVIDIA GeForce GTX 1080        Off | 00000000:25:00.0 Off |                  N/A |
|  0%   34C    P8               5W / 180W |      2MiB /  8192MiB |      0%      Default |
|                                         |                      |                  N/A |
+-----------------------------------------+----------------------+----------------------+

+---------------------------------------------------------------------------------------+
| Processes:                                                                            |
|  GPU   GI   CI        PID   Type   Process name                            GPU Memory |
|        ID   ID                                                             Usage      |
|=======================================================================================|
|  No running processes found                                                           |
+---------------------------------------------------------------------------------------+
  • Check for persistence: (if this is disabled or failed then this could be your issue)
sudo systemctl status nvidia-persistenced

Failed to initialize NVML: Failed to Initialize in Docker​

This error may occur inside of the emby docker container where it will not register nvidia-smi or any associated video cards. I reported this to linuxserver.io in their github as an issue here: docker-emby:#41.

Error:​

$ docker exec -it emby /bin/bash
root@5a15af76e670:/# nvidia-smi
Failed to initialize NVML: Unknown Error

Solution​

TLDR - solved this in Discord for anyone that comes across this in google and wants to try this solution.

The container wouldn't initialize the nvidia-smi and any cards associated with it. So mapping the nvidia stuff directly (as seen in the comment above in the docker command) is what ultimately solved it.

--device=/dev/nvidia-uvm \

--device=/dev/nvidia-uvm-tools
--device=/dev/nvidia-modeset
--device=/dev/nvidiactl
--device=/dev/nvidia0
--device=/dev/nvidia1 \

In my docker compose it looks like this:

emby:
  image: lscr.io/linuxserver/emby:latest
  container_name: emby
  environment:
    - PUID=${EMBY_UID}
    - PGID=${EMBY_GID}
    - NVIDIA_VISIBLE_DEVICES=all
    - NVIDIA_DRIVER_CAPABILITIES=all
    - TZ=${TIMEZONE}
  volumes:
    - ${DEFAULT_CONTAINER_DATA_LOCATION}/Emby:/config # Configuration directory
    - ./images/logowhite.png:/app/emby/system/dashboard-ui/modules/logoscreensaver/logowhite.png
    - ./images/logowhite.png:/app/emby/system/dashboard-ui/modules/themes/logowhite.png
    - ./images/logodark.png:/app/emby/system/dashboard-ui/modules/themes/logodark.png
    - ${LOCAL_TV_PATH}:/media/Synology/Television # Media directory
    - ${LOCAL_MOVIES_PATH}:/media/Synology/Movies # Media directory
    - ${LOCAL_BACKUPS_PATH}:/media/Synology/Backups #Backups Directory
    - /ssl/fullchain.pem:/ssl/fullchain.pem
    - /ssl/privkey.pem:/ssl/privkey.pem
    - /ssl/token:/ssl/token
  ports:
    - ${EMBY_HOST_PORT}:8096 #http port
    - ${EMBY_HOST_PORT_SSL}:8920 #ssl port
  runtime: nvidia
  restart: unless-stopped
  devices:
    - /dev/nvidia-uvm:/dev/nvidia-uvm # Added nvidia devices here
    - /dev/nvidia-uvm-tools:/dev/nvidia-uvm-tools # Added nvidia devices here
    - /dev/nvidia-modeset:/dev/nvidia-modeset # Added nvidia devices here
    - /dev/nvidiactl:/dev/nvidiactl # Added nvidia devices here
    - /dev/nvidia0:/dev/nvidia0 # Added nvidia devices here
    - /dev/nvidia1:/dev/nvidia1 # Added nvidia devices here (i have a 2nd GPU so i needed this too)
    - /dev/dri:/dev/dri # I added this per suggestion, but this is for VAAPI so i don't know if this actually works -- this was what was failing before.
  profiles:
    - emby

Screenshots​

nvidia-smi in the container:

Screenshot 2024-04-07 at 12 01 21 PM

Inside of emby in the transcoding section:

Screenshot 2024-04-07 at 12 01 41 PM