Can't get it to detect Nvidia GPU (Docker)

[cy1der]

I'm on Debian 13, driver version 550.163.01 and CUDA version 12.4 (latest available in packages).

Compose:

vertd:
    container_name: vertd
    image: ghcr.io/vert-sh/vertd:latest
    environment:
      PUBLIC_URL: https://vt.mydomain.tld/d
      ADMIN_PASSWORD: xxx
      NVIDIA_DISABLE_REQUIRE: 1
      NVIDIA_DRIVER_CAPABILITIES: all
      NVIDIA_VISIBLE_DEVICES: all
      # WEBHOOK_PINGS:
      # WEBHOOK_URL:
    ports:
      - 24153:24153
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]

I've tried bypassing the CUDA requirement with the nvidia environment variables (speaking of, why 12.8?) to avoid this error message when starting the container: unsatisfied condition: cuda>=12.8, please update your driver to a newer version, or use an earlier cuda container

Logs:

vertd  | [2025-12-10T01:58:31Z INFO  vertd] starting vertd
vertd  | [2025-12-10T01:58:31Z INFO  vertd] working w/ ffmpeg 6.1.1-3ubuntu5 and ffprobe 6.1.1-3ubuntu5
vertd  | [2025-12-10T01:58:31Z WARN  vertd::converter::gpu] *******
vertd  | [2025-12-10T01:58:31Z WARN  vertd::converter::gpu] you're running vertd on a docker container, but no GPU was detected.
vertd  | [2025-12-10T01:58:31Z WARN  vertd::converter::gpu] this usually is because you're running Docker under WSL or because
vertd  | [2025-12-10T01:58:31Z WARN  vertd::converter::gpu] you are not passing the GPU device correctly.
vertd  | [2025-12-10T01:58:31Z WARN  vertd::converter::gpu] 
vertd  | [2025-12-10T01:58:31Z WARN  vertd::converter::gpu] if this doesn't seem right, make sure to provide the following info when
vertd  | [2025-12-10T01:58:31Z WARN  vertd::converter::gpu] asking for help:
vertd  | [2025-12-10T01:58:31Z WARN  vertd::converter::gpu] - adapter name: llvmpipe (LLVM 20.1.2, 256 bits)
vertd  | [2025-12-10T01:58:31Z WARN  vertd::converter::gpu] - adapter vendor: 0x10005
vertd  | [2025-12-10T01:58:31Z WARN  vertd::converter::gpu] - backend: vulkan
vertd  | [2025-12-10T01:58:31Z WARN  vertd::converter::gpu] - device ID: 0
vertd  | [2025-12-10T01:58:31Z WARN  vertd::converter::gpu] - device type: Cpu
vertd  | [2025-12-10T01:58:31Z WARN  vertd::converter::gpu] - driver: llvmpipe
vertd  | [2025-12-10T01:58:31Z WARN  vertd::converter::gpu] - driver info: Mesa 25.0.7-0ubuntu0.24.04.2 (LLVM 20.1.2)
vertd  | [2025-12-10T01:58:31Z WARN  vertd::converter::gpu] 
vertd  | [2025-12-10T01:58:31Z WARN  vertd::converter::gpu] vertd will fall back to CPU rendering to ensure conversions can still proceed.
vertd  | [2025-12-10T01:58:31Z WARN  vertd::converter::gpu] *******
vertd  | [2025-12-10T01:58:31Z INFO  vertd] using CPU rendering (software encoding) -- this will be slower than GPU acceleration
vertd  | [2025-12-10T01:58:31Z INFO  vertd::http] http server listening on 0.0.0.0:24153

Am I doing something wrong?

[azurejelly]

try setting VERTD_FORCE_GPU to nvidia: https://github.com/VERT-sh/vertd/blob/main/docs/DOCKER_SETUP.md#manual-gpu-selection

I've tried bypassing the CUDA requirement with the nvidia environment variables (speaking of, why 12.8?) to avoid this error message when starting the container

we could look into downgrading to an older cuda image

[cy1der]

try setting VERTD_FORCE_GPU to nvidia

Here are the logs after going to Vert and uploading video to test:

vertd  | [2025-12-10T21:53:46Z INFO  vertd] starting vertd
vertd  | [2025-12-10T21:53:46Z INFO  vertd] working w/ ffmpeg 6.1.1-3ubuntu5 and ffprobe 6.1.1-3ubuntu5
vertd  | [2025-12-10T21:53:46Z INFO  vertd] using GPU from environment variable VERTD_FORCE_GPU: NVIDIA
vertd  | [2025-12-10T21:53:46Z INFO  vertd] detected a NVIDIA GPU -- if this isn't your vendor, open an issue.
vertd  | [2025-12-10T21:53:46Z INFO  vertd::http] http server listening on 0.0.0.0:24153
vertd  | [2025-12-10T21:54:00Z INFO  vertd::http::services::upload] uploaded file: Replay 2025-07-02 22-33-14.mov
vertd  | [2025-12-10T21:54:01Z INFO  vertd::converter] running 'ffmpeg -hide_banner -loglevel error -progress pipe:1 -hwaccel cuda -i input/5a19e00e-4e8d-452d-bb42-c576bc75652e.mov -c:v h264_nvenc -c:a aac -strict experimental -preset medium -b:v 23336859 -map_metadata 0 -map_chapters 0 output/5a19e00e-4e8d-452d-bb42-c576bc75652e.mkv'
vertd  | [2025-12-10T21:54:01Z ERROR vertd::converter] [AVHWDeviceContext @ 0x564528ed93c0] cu->cuInit(0) failed -> CUDA_ERROR_COMPAT_NOT_SUPPORTED_ON_DEVICE: forward compatibility was attempted on non supported HW
vertd  | [2025-12-10T21:54:01Z ERROR vertd::converter] Device creation failed: -542398533.
vertd  | [2025-12-10T21:54:01Z ERROR vertd::converter] No device available for decoder: device type cuda needed for codec hevc.
vertd  | [2025-12-10T21:54:01Z ERROR vertd::converter] [vist#0:0/hevc @ 0x564528d3e600] Hardware device setup failed for decoder: Generic error in an external library
vertd  | [2025-12-10T21:54:01Z ERROR vertd::converter] [vost#0:0/h264_nvenc @ 0x564528d2aa40] Error initializing a simple filtergraph
vertd  | [2025-12-10T21:54:01Z ERROR vertd::converter] Error opening output file output/5a19e00e-4e8d-452d-bb42-c576bc75652e.mkv.
vertd  | [2025-12-10T21:54:01Z ERROR vertd::converter] Error opening output files: Generic error in an external library
vertd  | [2025-12-10T21:54:01Z ERROR vertd::http::services::websocket] job 5a19e00e-4e8d-452d-bb42-c576bc75652e failed
vertd  | [2025-12-10T21:54:01Z ERROR vertd::http::services::websocket] failed to handle job failure: environment variable not found

[JovannMC]

in the dockerfile, try changing L9 to FROM nvidia/cuda:12.4.0-base-ubuntu20.04 to see if it helps