Lost GPU support after Home assistant Operating system update v17

[frixo3190]

Hi

I just migrate to the v17 version of HA Operating system, and I've launch the driver docker migration.

After that. I notice my Frigate module stop using my GPU !

Any idea ?

AVHWDeviceContext @

Ox5570007db840] libva: /usr/lib/

x86_64-linux-gnu/dri/iHD_drv_video.so

init failed

[inkeep-ai-assistant[bot]]

The error libva: /usr/lib/x86_64-linux-gnu/dri/iHD_drv_video.so init failed indicates that the Intel iHD driver is failing to initialize for hardware acceleration(1).

After a Home Assistant OS update, you may need to verify your hardware acceleration configuration. For Intel-based systems, the default driver is iHD, but you may need to change it to i965 depending on your CPU generation(1).

Try switching to the i965 driver:

Add the following environment variable to your Frigate configuration. For Home Assistant Add-on users, you can add this in the config.yml(1):

environment_vars:
  LIBVA_DRIVER_NAME: i965

(1)

Verify your hwaccel configuration:

Make sure you're using the appropriate preset for Intel hardware. For VAAPI (which works with both H.264 and H.265 streams automatically), use(1):

ffmpeg:
  hwaccel_args: preset-vaapi

(1)

Check device access:

Ensure the /dev/dri/renderD128 device is accessible to the Frigate container. If running through Docker, you need to pass this device through(1).

CPU Generation matters:

According to the documentation, different Intel CPU generations work better with different drivers(1):

• Gen 1-5: Use i965 driver with preset-vaapi
• Gen 6-7: Use iHD driver with preset-vaapi
• Gen 8-12: Use iHD driver (can use preset-vaapi or preset-intel-qsv-*)
• Gen 13+: Use iHD/Xe driver with preset-intel-qsv-*

Check your Intel CPU generation to determine the correct driver and preset(1).

If switching to i965 doesn't resolve the issue, check the Frigate logs for additional error details and verify that your hardware acceleration device is properly passed through to the container after the OS update.

📚 Sources:

1. Video Decoding

---

Have another question? Just tag @inkeep.

[frixo3190]

This is my conf
I've try many things, no way to get back GPU !

using the HomeAssistant Module
I've try the FullAccess Module too, no way !

environment_vars:
  LIBVA_DRIVER_NAME: i965
mqtt:
  enabled: true
  host: xxxxxxxxxx
  port: 1883
  user: xxxxxxxxxx
  password: xxxxxxxxxx
  #topic_prefix: frigate
logger:
  default: debug
go2rtc:
  log:
    level: debug
    api: debug
    exec: debug
    rtsp: debug
    streams: debug
    webrtc: debug
  streams:
    terrasse_cuisine:
      - rtsp://xxxxxxxxxx
    terrasse_cuisine_hd:
      - rtsp://xxxxxxxxxxxx
    cuisine:
      - rtsp://xxxxxxxxxx
      #- ffmpeg:cuisine#audio=opus
    cuisine_hd:
      - rtsp://xxxxxxxxxx
      - ffmpeg:cuisine_hd#audio=opus
    salon:
      - rtsp://xxxxxxxxxx
      #- ffmpeg:salon#audio=opus
    salon_hd:
      - rtsp://xxxxxxxxxx
      #- ffmpeg:salon_hd#audio=opus
    portail_hd:
      - rtsp://xxxxxxxxxx
      #- ffmpeg:portail_hd#audio=opus
    potager:
      - rtsp://xxxxxxxxxx
      #- ffmpeg:Potager#audio=opus
    potager_hd:
      - rtsp://xxxxxxxxxx
      - ffmpeg:Potager#audio=opus
    celier_hd:
      - rtsp://xxxxxxxxxx
      - ffmpeg:celier_hd#audio=opus
    celier:
      - rtsp://xxxxxxxxxx
      #- ffmpeg:celier#audio=opus
    ruisseau_hd:
      - rtsp://xxxxxxxxxx
      - ffmpeg:ruisseau_hd#audio=opus
    ruisseau:
      - rtsp://xxxxxxxxxx
      #- ffmpeg:ruisseau#audio=opus
# Optional: Set the encode quality of the jsmpeg stream (default: shown below)
# 1 is the highest quality, and 31 is the lowest. Lower quality feeds utilize less CPU resources.
live:
  quality: 8
ffmpeg:
  #hwaccel_args: -hwaccel vaapi -hwaccel_device /dev/dri/renderD128
  hwaccel_args: preset-vaapi
  retry_interval: 10
detectors:
  # detector_name:
  #   type: cpu
  coral:
    type: edgetpu
    device: usb
  # ov:
  #   type: openvino
  #   device: AUTO
  #   model:
  #     path: /openvino-model/ssdlite_mobilenet_v2.xml
objects:
  # Optional: list of objects to track from labelmap.txt (default: shown below)
  track:
    - person
    - car
    - motorcycle
    - bicycle
  #filters:
  #car:
  #mask:
  #  - 1038,720,729,426,92,426,0,515,0,720
  # person:
  #   mask:
  #     - 0.152,0.274,0.317,0.217,0.319,0.129,0.309,0.067,0.246,0.07,0.19,0.112,0.154,0.142
# Include all cameras by default in Birdseye view
birdseye:
  enabled: false
  restream: true
review:
  alerts:
    labels:
      - person
      - car
      - motorcycle
      - bicycle
record:
  enabled: true
  retain:
    days: 2
  alerts:
    retain:
      days: 3
  detections:
    retain:
      days: 3
snapshots:
  enabled: true
  retain:
    default: 20
cameras:
  portail:
    enabled: true
    ffmpeg:
      output_args:
        record: preset-record-generic
      inputs:
        - path: rtsp://localhost:8554/portail_hd
          input_args: preset-rtsp-restream
          roles:
            - record
            # - path: rtsp://localhost:8554/portail # Maintenant ce flux marchera via Go2RTC
            #   input_args: preset-rtsp-restream
            #   roles:
            #- detect
            #Ne fonctionne pas
            # - path: rtsp://localhost:8554/portail
            #   input_args: preset-rtsp-restream #preset-rtsp-restream
            #   roles:
            - detect
    live:
      streams:
        HD: portail_hd
        #SD: portail
    ui:
      order: 0
      dashboard: true
    detect:
      enabled: true
      width: 640
      height: 360
      fps: 5
    motion:
      threshold: 30
      contour_area: 14
      improve_contrast: true
  terrasse_cuisine:
    enabled: true
    ffmpeg:
      output_args:
        record: preset-record-generic
      inputs:
        - path: rtsp://localhost:8554/terrasse_cuisine_hd
          input_args: preset-rtsp-restream
          roles:
            - record
        - path: rtsp://localhost:8554/terrasse_cuisine
          input_args: preset-rtsp-restream
          roles:
            - detect
    live:
      streams:
        HD: terrasse_cuisine_hd
        SD: terrasse_cuisine
    ui:
      order: 1
      dashboard: true
    detect:
      enabled: true
      width: 640
      height: 360
      fps: 5
  celier:
    enabled: true
    ffmpeg:
      output_args:
        record: preset-record-generic
      inputs:
        - path: rtsp://localhost:8554/celier_hd
          input_args: preset-rtsp-restream #preset-rtsp-generic
          roles:
            - record
        - path: rtsp://localhost:8554/celier
          input_args: preset-rtsp-restream
          roles:
            - detect
    live:
      streams:
        HD: celier_hd
        SD: celier
    ui:
      order: 2
      dashboard: true
    detect:
      enabled: true
      width: 640
      height: 360
      fps: 5
    motion:
      threshold: 30
      contour_area: 20
      improve_contrast: true
  cuisine:
    enabled: true
    ffmpeg:
      output_args:
        record: preset-record-generic
      inputs:
        - path: rtsp://localhost:8554/cuisine_hd
          input_args: preset-rtsp-restream
          roles:
            - record
        - path: rtsp://localhost:8554/cuisine
          input_args: preset-rtsp-restream
          roles:
            - detect
    live:
      streams:
        HD: cuisine_hd
        SD: cuisine
    ui:
      order: 3
      dashboard: true
    detect:
      enabled: true
      width: 640
      height: 360
      fps: 5
    motion:
      threshold: 30
      contour_area: 20
      improve_contrast: true
  salon:
    enabled: true
    ffmpeg:
      output_args:
        record: preset-record-generic
      inputs:
        - path: rtsp://localhost:8554/salon
          input_args: preset-rtsp-restream
          roles:
            - detect
        - path: rtsp://localhost:8554/salon_hd
          input_args: preset-rtsp-restream
          roles:
            - record
    live:
      streams:
        HD: salon_hd
        SD: salon
    ui:
      order: 4
      dashboard: true
    objects:
      filters:
        person: {}
    motion:
      threshold: 30
      contour_area: 25
      improve_contrast: true
  ruisseau:
    enabled: true
    ffmpeg:
      output_args:
        record: preset-record-generic
      inputs:
        - path: rtsp://localhost:8554/ruisseau_hd
          input_args: preset-rtsp-restream
          roles:
            - record
        - path: rtsp://localhost:8554/ruisseau
          input_args: preset-rtsp-restream
          roles:
            - detect
    live:
      streams:
        HD: ruisseau_hd
        SD: ruisseau
    ui:
      order: 5
      dashboard: true
    detect:
      enabled: true
      width: 640
      height: 360
    motion:
      threshold: 39
      contour_area: 35
      improve_contrast: true
  potager:
    enabled: true
    ffmpeg:
      output_args:
        record: preset-record-generic
      inputs:
        - path: rtsp://localhost:8554/potager_hd
          input_args: preset-rtsp-restream
          roles:
            - detect
            # - path: rtsp://localhost:8554/potager_hd
            #   input_args: preset-rtsp-restream
            #   roles:
            - record
    live:
      streams:
        HD: potager_hd
        #SD: potager
    motion:
      threshold: 30
      contour_area: 21
      improve_contrast: true
    detect:
      enabled: false
    ui:
      order: 6
      dashboard: true
version: 0.16-0
camera_groups:
  test:
    order: 1
    icon: LuActivitySquare
    cameras:
      - birdseye
      - celier
      - potager
      - ruisseau
      - cuisine
      - portail
      - salon
detect:
  enabled: true

[NickM-27]

It's saying the driver isn't installed. You'll need to seek support from the OS

[frixo3190]

Having a i5 4th génération, I must use the 'i965' driver. It must be auto with the preset-vaapi args.
If I try to force "hwaccel_args: -hwaccel vaapi -hwaccel_device /dev/dri/renderD128" nothing happen (before the upgrade v17 I was using that : hwaccel_args: -hwaccel vaapi -hwaccel_device /dev/dri/renderD128)

[mainmind83]

Same issue — Intel N5105 (Jasper Lake), HAOS 17.3 bare metal, Frigate Full Access 0.17.1

Previously working with VAAPI hardware acceleration. After updating to HAOS 17.3, /dev/dri/renderD128 exists on the host but is not passed through to the Frigate container.

Environment

• Hardware: Intel Celeron N5105 (Jasper Lake, Gen11 iGPU)
• HAOS: 17.3 bare metal
• Frigate add-on: Full Access 0.17.1 (Frigate 0.17.1-416a9b7)

Confirmed behavior

• Host: ls /dev/dri/ shows card0 and renderD128 ✅
• Container: /dev/dri/renderD128 not visible inside the Frigate Full Access add-on ❌

Tried without success

• hwaccel_args: preset-vaapi
• Manual hwaccel args
• environment_vars: LIBVA_DRIVER_NAME: iHD
• OpenVINO detector with device: GPU

All fail with the same errors:

libva: /usr/lib/x86_64-linux-gnu/dri/iHD_drv_video.so init failed
libva: /usr/lib/x86_64-linux-gnu/dri/i965_drv_video.so init failed
Failed to initialise VAAPI connection: -1 (unknown libva error)
Device creation failed: -5

OpenVINO GPU error:

[GPU] Context was not initialized for 0 device

Current workaround

hwaccel_args: [] + OpenVINO device: AUTO (falls back to CPU). System functional but iGPU completely unused, causing sustained ~25% CPU load for a single 704x576@8fps stream.

This appears to be a regression in HAOS 17.x breaking /dev/dri device passthrough to add-on containers. Same hardware worked fine on previous HAOS versions.

Additional note

Switched from standard Frigate NVR add-on to Frigate NVR Full Access specifically to enable GPU access, as the standard add-on has Protection Mode enabled by default. Even with Full Access, /dev/dri/renderD128 is not passed through to the container.

With the standard Frigate add-on, the Google Coral USB detector worked correctly — suggesting USB devices are properly passed through, but DRI devices are not.