Is there an existing issue for this?
Current Behavior
After pulling the current lscr.io/linuxserver/jellyfin:10.11.11, Intel Quick Sync fails to initialize on an Intel Arc A310.
VA-API opens successfully; creating QSV from that VA-API device fails:
Error setting child device handle: -17
Device creation failed: -1313558101.
Failed to set value 'qsv=qs@va' for option 'init_hw_device': Unknown error occurred
A previous local image that used to be tagged 10.11.11 (now untagged / retagged as 10.11.11-previous) works correctly with the same host, devices, and settings. Full hardware transcodes complete at ~40× realtime with hevc_qsv.
This looks like an in-place rebuild of the 10.11.11 tag that changed the jellyfin-ffmpeg / Intel media stack.
Expected Behavior
QSV derived from VA-API (-init_hw_device vaapi=... -init_hw_device qsv=qs@va) should initialize and transcode on Arc A310, as it did with the previous 10.11.11 image build.
Steps To Reproduce
-
Host: TrueNAS Scale, dual Intel Arc A310; container only gets one card:
/dev/dri/card2
/dev/dri/renderD129
-
Run current image: lscr.io/linuxserver/jellyfin:10.11.11
-
Inside the container:
/usr/lib/jellyfin-ffmpeg/ffmpeg -hide_banner \
-init_hw_device vaapi=va:/dev/dri/renderD129,driver=iHD \
-init_hw_device qsv=qs@va \
-f lavfi -i nullsrc=s=640x360:d=1 -frames:v 1 -f null -
- Observe failure with -17 / qsv=qs@va.
- Switch to previous image (local ID 54e2a6bc7d26, digest below) with the same devices/config.
- Same command succeeds; Jellyfin QSV transcodes work (SDR HEVC → hevc_qsv).
VA-API-only init succeeds on both images:
/usr/lib/jellyfin-ffmpeg/ffmpeg -hide_banner \
-init_hw_device vaapi=va:/dev/dri/renderD129,driver=iHD \
-f lavfi -i nullsrc=s=640x360:d=1 -frames:v 1 -f null -
Environment
OS: TrueNAS Scale 25.10.x (Linux)
CPU architecture: x86-64
GPU: Intel Arc A310 (Alchemist)
Docker: TrueNAS apps / compose
Devices passed: single Arc render node (renderD129 + card2)
HW acceleration: Intel QSV (Jellyfin); failure is at ffmpeg device init, independent of UI
Role,Tag / note,Image ID,Repo digest
Broken (current Hub tag),10.11.11,06822bf918e5,sha256:b8dcc7b71d0ea872b74314da4b995c0cf282b1778438c295996e7be88c70fdda
Working (previous build),"was 10.11.11, now local only",54e2a6bc7d26,sha256:bb8d372e35d5c4a6cb61d830a06f5b5846528315b97cf5d38b80eea1e430efa7
docker buildx imagetools inspect lscr.io/linuxserver/jellyfin:10.11.11 matches the broken index digest b8dcc7b7….
ffmpeg version difference
Working image: jellyfin-ffmpeg built with gcc 13 (Ubuntu 24.04)
Broken image: jellyfin-ffmpeg built with gcc 15 (Ubuntu 15.x / newer base)
Both report ffmpeg version 7.1.4-Jellyfin, but the build environment differs.
CPU architecture
x86-64
Docker creation
Compose
jellyfin:
container_name: jellyfin
image: lscr.io/linuxserver/jellyfin:${JELLYFIN_VERSION}
hostname: jellyfin
restart: unless-stopped
devices:
- ${ARC_CARD_2}
- ${ARC_RENDER_2}
environment:
- PUID=1000
- PGID=4000
- TZ=${TZ}
volumes:
- ${JELLYFIN_CONFIG}:/config
- ${JELLYFIN_MEDIA}:/data/media:ro
- ${TESTMEDIALIB}:/data/testmedialib:ro
tmpfs:
- /tmptranscode:uid=1000,gid=4000,mode=1777,size=25g
networks:
proxy:
ipv4_address: 172.24.1.100
expose:
- 8096
labels:
- traefik.enable=true
- traefik.http.routers.jellyfin.entrypoints=websecure
- traefik.http.routers.jellyfin.rule=Host(`${JELLYFIN_DN}`)
- traefik.http.routers.jellyfin.tls=true
- traefik.http.services.jellyfin.loadbalancer.server.port=8096
cpus: 8
mem_limit: 32g
Container logs
libva info: va_openDriver() returns 0
[AVHWDeviceContext] Error setting child device handle: -17
Device creation failed: -1313558101.
Failed to set value 'qsv=qs@va' for option 'init_hw_device': Unknown error occurred
Is there an existing issue for this?
Current Behavior
After pulling the current lscr.io/linuxserver/jellyfin:10.11.11, Intel Quick Sync fails to initialize on an Intel Arc A310.
VA-API opens successfully; creating QSV from that VA-API device fails:
A previous local image that used to be tagged 10.11.11 (now untagged / retagged as 10.11.11-previous) works correctly with the same host, devices, and settings. Full hardware transcodes complete at ~40× realtime with hevc_qsv.
This looks like an in-place rebuild of the 10.11.11 tag that changed the jellyfin-ffmpeg / Intel media stack.
Expected Behavior
QSV derived from VA-API (-init_hw_device vaapi=... -init_hw_device qsv=qs@va) should initialize and transcode on Arc A310, as it did with the previous 10.11.11 image build.
Steps To Reproduce
Host: TrueNAS Scale, dual Intel Arc A310; container only gets one card:
/dev/dri/card2
/dev/dri/renderD129
Run current image: lscr.io/linuxserver/jellyfin:10.11.11
Inside the container:
VA-API-only init succeeds on both images:
Environment
CPU architecture
x86-64
Docker creation
Compose jellyfin: container_name: jellyfin image: lscr.io/linuxserver/jellyfin:${JELLYFIN_VERSION} hostname: jellyfin restart: unless-stopped devices: - ${ARC_CARD_2} - ${ARC_RENDER_2} environment: - PUID=1000 - PGID=4000 - TZ=${TZ} volumes: - ${JELLYFIN_CONFIG}:/config - ${JELLYFIN_MEDIA}:/data/media:ro - ${TESTMEDIALIB}:/data/testmedialib:ro tmpfs: - /tmptranscode:uid=1000,gid=4000,mode=1777,size=25g networks: proxy: ipv4_address: 172.24.1.100 expose: - 8096 labels: - traefik.enable=true - traefik.http.routers.jellyfin.entrypoints=websecure - traefik.http.routers.jellyfin.rule=Host(`${JELLYFIN_DN}`) - traefik.http.routers.jellyfin.tls=true - traefik.http.services.jellyfin.loadbalancer.server.port=8096 cpus: 8 mem_limit: 32gContainer logs