GPU Passthrough for Proxmox LXCs#

What Was Established#

• Intel GPU passthrough to Proxmox LXCs requires both host-side module loading and specific LXC device mounts.
• vainfo frequently fails with X11/X server errors in headless containers; this is expected and does not indicate a broken passthrough.
• intel_gpu_top requires i915 module loaded on the host and accessible debugfs/sysfs paths inside the container.

Key Decisions#

• Headless VA-API Testing: When vainfo reports error: can't connect to X server!, set environment variables to bypass X11:

  export XDG_RUNTIME_DIR=/tmp/runtime-root
  export LIBVA_DRIVER_NAME=iHD
  vainfo

• Module Verification: lsmod | grep i915 confirms the driver is loaded inside the container. Presence of i915, drm_buddy, ttm, and drm_display_helper indicates successful module injection.
• Device Access: intel_gpu_top failing with No device filter specified... typically points to missing debugfs mounts or host-side i915 parameters, not necessarily a broken /dev/dri/ passthrough.

Current Configuration#

Proxmox LXC Config (/etc/pve/lxc/<container-id>.conf):

Jellyfin LXC GPU Passthrough & Hardware Acceleration#

What Was Established#

Successfully configured Intel UHD Graphics 630 GPU passthrough to a Jellyfin LXC container on Proxmox for hardware-accelerated transcoding via Intel QuickSync (QSV).

Key Decisions#

• GPU Passthrough Method: LXC container-level GPU device mapping (not full VM passthrough)
• Hardware Acceleration: Intel QuickSync (QSV) selected over VAAPI for Jellyfin’s native support
• Monitoring Constraints: Accepted that LXC container restrictions prevent full GPU monitoring tools (dmesg, intel_gpu_top) from functioning; validated functionality through actual transcoding tests instead

Current Configuration#

Host GPU Details#

• GPU: Intel Corporation CoffeeLake-S GT2 [UHD Graphics 630]
• PCI Address: 00:02.0
• Driver: i915 (loaded)
• Related Modules: drm_buddy, ttm, drm_display_helper, cec, i2c_algo_bit, video

LXC Container GPU Devices#

• /dev/dri/card0 — character special (major 226, minor 0)
• /dev/dri/renderD128 — character special (major 226, minor 128)
• Permissions: crw-rw---- root:video (226/0 and 226/128)

Jellyfin Configuration#

1. User Group Assignment: jellyfin user added to video group (usermod -a -G video jellyfin)
2. Dashboard → Playback Settings:
   • Hardware Acceleration: Intel QuickSync (QSV)
   • Enable hardware encoding: Yes
   • Enable hardware decoding: Yes
   • Enable tone mapping: Yes
   • Allow encoding in HEVC: Yes
   • Allow encoding in AV1: Yes (if supported)

Validation Commands#

# Verify GPU device accessibility
ls -la /dev/dri/

# Check if processes are using GPU during playback
lsof /dev/dri/renderD128

# Monitor Jellyfin logs for hardware acceleration
sudo journalctl -u jellyfin -f | grep -i "hardware\|qsv\|quicksync\|vaapi"

# Check active transcoding sessions in Jellyfin UI
# Dashboard → Active Devices → look for (HW) indicator

Historical Notes#

• dmesg restriction: dmesg: read kernel buffer failed: Operation not permitted — expected in LXC containers without full device access
• intel_gpu_top limitation: GPU monitoring tools that require kernel debugfs access will not work inside the LXC; validated via actual transcoding performance and log inspection instead
• i915 driver loaded: Confirmed via lsmod | grep i915 showing 3,928,064 bytes loaded
• No GPU debug info: /sys/kernel/debug/dri not available in container — accepted limitation

Open Questions#

• Does AV1 hardware encoding actually work on Coffee Lake-S (Gen 9.5) — typically limited to H.264/H.265
• Performance baseline: what CPU load reduction is observed with QSV vs software transcoding?
• Can GPU passthrough be extended to other LXC containers (e.g., Plex, if migrated)

Related Pages#

• Plex Transcoding LXC — similar GPU passthrough patterns for Plex
• Proxmox LXC Configuration — LXC container setup patterns
• Servarr - Media Automation Stack — media server ecosystem

Gluetun VPN Service