Jellyfin + AMD GPU

[Ivansallas]

Jellyfin + AMD GPU (VAAPI) works in Compose Service but fails in App Service

Environment

• EasyPanel: Latest version (June 2026)
• OS: Ubuntu Server 24.04
• GPU: AMD RX 580
• Driver: amdgpu
• Container image: linuxserver/jellyfin:latest

Problem

When deploying Jellyfin as a regular EasyPanel App (Docker Swarm service), the container receives the /dev/dri bind mount, but hardware acceleration does not work.

Inside the container:

ffmpeg -vaapi_device /dev/dri/renderD128

returns:

No VA display found for device /dev/dri/renderD128

and:

Operation not permitted

The service inspection shows:

"Devices": null,
"DeviceCgroupRules": null

even though /dev/dri is mounted.

Verification

The user abc belongs to:

video
render group

and permissions are correct.

Inside the container:

/dev/dri/card0
/dev/dri/renderD128

are visible.

However VAAPI initialization fails.

Workaround

Deploying the exact same Jellyfin configuration as a Compose Service works immediately.

Compose configuration:

devices:
  - /dev/dri:/dev/dri
group_add:
  - "44"
  - "992"

After migrating to Compose:

VAAPI device /dev/dri/renderD128 is AMD GPU
VAAPI device /dev/dri/renderD128 supports Vulkan DRM interop

FFmpeg starts correctly using:

-init_hw_device drm=dr:/dev/dri/renderD128
-init_hw_device vaapi=va@dr
-init_hw_device vulkan=vk@dr
-hwaccel vaapi
h264_vaapi

and hardware transcoding works.

Expected behavior

A standard EasyPanel App deployment should provide the same GPU access as a Compose Service when /dev/dri is configured.

Result

• App Service (Swarm): VAAPI fails.
• Compose Service: VAAPI works correctly.
• Same host.
• Same image.
• Same GPU.
• Same mounts.

This suggests a difference in how EasyPanel creates Swarm services versus Compose services regarding GPU device permissions and DRM access.