[HW Accel Support]: High Detector CPU Usage using Intel iGPU Beelink EQ12

[bigadz87]

Describe the problem you are having

Hi!

I'm running Frigate in docker on a Debian 13 LXC with passed through N100 iGPU (Intel Alderlake N Gen12), however Frigate only shows gpu usage for a little while if I restart the proxmox node and then shows 0.0% The detector CPU usage is always quite high (over 70%). I have 5 camera's typically running (2 x Reolink Duos, 1 x Reolink RLC-1224a and 2 x Eufy 2k indoor cams) and I'm using OpenVino with YOLOv9-S.

I'm not sure if this is too much for the system or if I've done something wrong but would appreciate some guidance.

Version

0.16.1-e664cb2

Frigate config file

mqtt:
  host: 192.168.1.233
  port: 1883
  user: mqtt-user
  password: *****

go2rtc:
  streams:
    Reolink_Front_Door:
      - rtsp://user:[email protected]:554/h265Preview_01_main
      #- ffmpeg:Reolink_Front_Door#audio=aac
    Reolink_Front_Door_Sub:
      - rtsp://user:[email protected]:554/Preview_01_sub
      #- ffmpeg:Reolink_Front_Door_Sub#audio=aac
    Reolink_Back_Door:
      - rtsp://user:[email protected]:554/h265Preview_01_main
      #- ffmpeg:Reolink_Back_Door#audio=aac
    Reolink_Back_Door_Sub:
      - rtsp://user:[email protected]:554/Preview_01_sub
      #- ffmpeg:Reolink_Back_Door_Sub#audio=aac
   # Reolink_Side_Door:
     # - rtsp://user:[email protected]:554/h265Preview_01_main
   # Reolink_Side_Door_Sub:
     # - rtsp://user:[email protected]:554/Preview_01_sub
    Eufy_Lounge_Room:
      - rtsp://user:[email protected]/live0
      #- ffmpeg:Eufy_Lounge_Room#audio=aac
    Eufy_Bedroom:
      - rtsp://user:[email protected]/live0
      #- ffmpeg:Eufy_Bedroom#audio=aac

detectors:
  ov:
    type: openvino
    device: GPU

model:
  model_type: yolo-generic
  width: 320
  height: 320
  input_tensor: nchw
  input_dtype: float
  path: /config/yolov9-s-320.onnx
  labelmap_path: /labelmap/coco-80.txt

objects:
  track:
    - car
    - person
    - dog

audio:
  enabled: true
  listen:
    - bark

face_recognition:
  enabled: true
lpr:
  enabled: true

cameras:
  Reolink_Front_Door:
    ffmpeg:
      hwaccel_args: preset-vaapi
      output_args:
        record: preset-record-generic-audio-copy
      inputs:
        - path: rtsp://127.0.0.1:8554/Reolink_Front_Door
          input_args: preset-rtsp-restream
          roles:
            - record
            - audio
        - path: rtsp://127.0.0.1:8554/Reolink_Front_Door_Sub
          input_args: preset-rtsp-restream
          roles:
            - detect
    detect:
      enabled: true
    record:
      enabled: true
      alerts:
        pre_capture: 10
        post_capture: 8
    review:
      alerts:
        enabled: true
        labels:
          - person
      detections:
        enabled: false

  Reolink_Back_Door:
    ffmpeg:
      hwaccel_args: preset-vaapi
      output_args:
        record: preset-record-generic-audio-copy
      inputs:
        - path: rtsp://127.0.0.1:8554/Reolink_Back_Door
          input_args: preset-rtsp-restream
          roles:
            - record
            - audio
        - path: rtsp://127.0.0.1:8554/Reolink_Back_Door_Sub
          input_args: preset-rtsp-restream
          roles:
            - detect
    detect:
      enabled: true
    record:
      enabled: true
      alerts:
        pre_capture: 10
        post_capture: 8
    review:
      alerts:
        enabled: true
        labels:
          - person
      detections:
        enabled: false

#  Reolink_Side_Door:
#    ffmpeg:
#      hwaccel_args: preset-vaapi
#      output_args:
#        record: preset-record-generic-audio-copy
#      inputs:
#        - path: rtsp://127.0.0.1:8554/Reolink_Side_Door
#          input_args: preset-rtsp-restream
#          roles:
#            - record
#            - audio
#        - path: rtsp://127.0.0.1:8554/Reolink_Side_Door_Sub
#          input_args: preset-rtsp-restream
#          roles:
#            - detect
#    detect:
#      enabled: true
#    record:
#      enabled: true
#      alerts:
#        pre_capture: 10
#        post_capture: 8
#    review:
#      alerts:
#        enabled: true
#        labels:
#          - person
#      detections:
#        enabled: false

  Eufy_Lounge_Room:
    ffmpeg:
      hwaccel_args: preset-vaapi
      output_args:
        record: preset-record-generic-audio-copy
      inputs:
        - path: rtsp://127.0.0.1:8554/Eufy_Lounge_Room
          input_args: preset-rtsp-restream
          roles:
            - record
            - detect
            - audio
    detect:
      enabled: true

  Eufy_Bedroom:
    ffmpeg:
      hwaccel_args: preset-vaapi
      output_args:
        record: preset-record-generic-audio-copy
      inputs:
        - path: rtsp://127.0.0.1:8554/Eufy_Bedroom
          input_args: preset-rtsp-restream
          roles:
            - record
            - detect
            - audio
    detect:
      enabled: true
detect:
  enabled: true
version: 0.16-0

docker-compose file or Docker CLI command

services:

  frigate:
    container_name: frigate
    privileged: true
    restart: unless-stopped
    image: ghcr.io/blakeblackshear/frigate:stable
    cap_add:
      - CAP_PERFMON
    shm_size: "512mb" # update for your cameras based on calculation above
    devices:
      - /dev/bus/usb:/dev/bus/usb
      - /dev/dri/renderD128:/dev/dri/renderD128
      - /dev/dri/card0:/dev/dri/card0
    volumes:
      - /etc/localtime:/etc/localtime:ro
      - /opt/frigate/config/:/config
      - /cctv:/media/frigate
      - type: tmpfs # Optional: 1GB of memory, reduces SSD/SD Card wear
        target: /tmp/cache
        tmpfs:
          size: 1000000000
    ports:
      - "5000:5000"
      - "1935:1935" # RTMP feeds
      - "8554:8554" # RTSP feeds
    environment:
      FRIGATE_RTSP_PASSWORD: "pass"

Relevant Frigate log output

2025-09-28 22:06:28.587856717  [INFO] Preparing Frigate...
2025-09-28 22:06:29.160483833  [INFO] Starting Frigate...
2025-09-28 22:06:31.558114743  [2025-09-28 22:06:31] frigate.util.config            INFO    : Checking if frigate config needs migration...
2025-09-28 22:06:31.577111302  [2025-09-28 22:06:31] frigate.util.config            INFO    : frigate config does not need migration...
2025-09-28 22:06:31.784649057  [2025-09-28 22:06:31] frigate.util.services          INFO    : Automatically detected vaapi hwaccel for video decoding
2025-09-28 22:06:41.576178675  [2025-09-28 22:06:41] frigate.app                    INFO    : Starting Frigate (0.16.1-e664cb2)
2025-09-28 22:06:41.590990877  [2025-09-28 22:06:41] peewee_migrate.logs            INFO    : Starting migrations
2025-09-28 22:06:41.591479554  [2025-09-28 22:06:41] peewee_migrate.logs            INFO    : There is nothing to migrate
2025-09-28 22:06:41.610331161  [2025-09-28 22:06:41] frigate.app                    INFO    : Recording process started: 423
2025-09-28 22:06:41.611116582  [2025-09-28 22:06:41] frigate.app                    INFO    : Review process started: 436
2025-09-28 22:06:41.614481739  [2025-09-28 22:06:41] frigate.app                    INFO    : go2rtc process pid: 123
2025-09-28 22:06:41.630517442  [2025-09-28 22:06:41] detector.ov                    INFO    : Starting detection process: 448
2025-09-28 22:06:41.630930755  [2025-09-28 22:06:41] frigate.app                    INFO    : Embedding process started: 451
2025-09-28 22:06:41.655548068  [2025-09-28 22:06:41] frigate.app                    INFO    : Output process started: 475
2025-09-28 22:06:41.681827010  �[1;31m2025-09-28 22:06:41.678909850 [E:onnxruntime:Default, env.cc:234 ThreadMain] pthread_setaffinity_np failed for thread: 491, index: 2, mask: {3, }, error code: 22 error msg: Invalid argument. Specify the number of threads explicitly so the affinity is not set.�[m
2025-09-28 22:06:41.772754861  [2025-09-28 22:06:41] frigate.app                    INFO    : Camera processor started for Reolink_Front_Door: 511
2025-09-28 22:06:41.803142737  [2025-09-28 22:06:41] frigate.app                    INFO    : Camera processor started for Reolink_Back_Door: 513
2025-09-28 22:06:41.833810553  [2025-09-28 22:06:41] frigate.app                    INFO    : Camera processor started for Eufy_Lounge_Room: 535
2025-09-28 22:06:41.834516149  [2025-09-28 22:06:41] frigate.app                    INFO    : Camera processor started for Eufy_Bedroom: 564
2025-09-28 22:06:41.873053614  [2025-09-28 22:06:41] frigate.app                    INFO    : Capture process started for Reolink_Front_Door: 585
2025-09-28 22:06:41.904812239  [2025-09-28 22:06:41] frigate.app                    INFO    : Capture process started for Reolink_Back_Door: 598
2025-09-28 22:06:41.935522258  [2025-09-28 22:06:41] frigate.app                    INFO    : Capture process started for Eufy_Lounge_Room: 611
2025-09-28 22:06:41.947534384  �[1;31m2025-09-28 22:06:41.946761483 [E:onnxruntime:Default, env.cc:234 ThreadMain] pthread_setaffinity_np failed for thread: 623, index: 2, mask: {3, }, error code: 22 error msg: Invalid argument. Specify the number of threads explicitly so the affinity is not set.�[m
2025-09-28 22:06:41.987817633  [2025-09-28 22:06:41] frigate.app                    INFO    : Capture process started for Eufy_Bedroom: 628
2025-09-28 22:06:42.046061029  INFO: Created TensorFlow Lite XNNPACK delegate for CPU.
2025-09-28 22:06:42.055820645  �[1;31m2025-09-28 22:06:42.052994634 [E:onnxruntime:Default, env.cc:234 ThreadMain] pthread_setaffinity_np failed for thread: 643, index: 2, mask: {3, }, error code: 22 error msg: Invalid argument. Specify the number of threads explicitly so the affinity is not set.�[m
2025-09-28 22:06:42.245366668  [2025-09-28 22:06:42] frigate.audio_manager          INFO    : Audio processor started (pid: 630)
2025-09-28 22:06:42.321361790  [2025-09-28 22:06:42] frigate.api.fastapi_app        INFO    : Starting FastAPI app
2025-09-28 22:06:42.501042146  �[1;31m2025-09-28 22:06:42.500593837 [E:onnxruntime:Default, env.cc:234 ThreadMain] pthread_setaffinity_np failed for thread: 728, index: 2, mask: {3, }, error code: 22 error msg: Invalid argument. Specify the number of threads explicitly so the affinity is not set.�[m
2025-09-28 22:06:42.711458386  loading data from : /config/model_cache/facedet/landmarkdet.yaml
2025-09-28 22:06:42.783305045  [2025-09-28 22:06:42] frigate.api.fastapi_app        INFO    : FastAPI started
2025-09-28 22:06:43.369829344  INFO: Created TensorFlow Lite XNNPACK delegate for CPU.

Relevant go2rtc log output

2025-09-28 22:06:29.156843978  [INFO] Preparing new go2rtc config...
2025-09-28 22:06:29.647627160  [INFO] Starting go2rtc...
2025-09-28 22:06:29.726479336  22:06:29.726 INF go2rtc platform=linux/amd64 revision=fa580c5 version=1.9.9
2025-09-28 22:06:29.726658209  22:06:29.726 INF config path=/dev/shm/go2rtc.yaml
2025-09-28 22:06:29.727168064  22:06:29.727 INF [rtsp] listen addr=:8554
2025-09-28 22:06:29.727358200  22:06:29.727 INF [api] listen addr=:1984
2025-09-28 22:06:29.727874823  22:06:29.727 INF [webrtc] listen addr=:8555
2025-09-28 22:06:38.465638930  [INFO] Starting go2rtc healthcheck service...

FFprobe output from your camera

n/a

Install method

Proxmox via Docker

Object Detector

OpenVino

Network connection

Mixed

Camera make and model

Reolink and Eufy

Screenshots of the Frigate UI's System metrics pages

Any other information that may be helpful

No response

[hawkeye217]

Your inference time, while slightly high, does suggest that the GPU is being used for inference. CPU inference would be much slower.

Detector CPU Usage is measuring the CPU required to prepare inputs for inference on the GPU. This task can't be offloaded to the GPU.

I'd suggest tuning your motion detection. If your cameras are constantly detecting motion, for example, object detection will constantly run as Frigate looks for objects in areas of motion. This is one common explanation for a high Detector CPU Usage value.

[NickM-27]

The CPU usage for the detector can be a bit high when there is a lot of activity, this is something that will be improved in the next version of Frigate (0.17)

You should use QSV instead of VAAPI for the cameras that are having errors

[bigadz87]

Ok I can change that, I use the h265 main stream and the h264 sub stream on the ReoLink cameras, do I need to set a different qsv version for each stream?

[hawkeye217]

Hardware acceleration is only used for video decoding on your detect stream. The record stream is not decoded.

[bigadz87]

Ok so to update it would be just adding preset-intel-qsv-h264 to the sub stream? Can I remove the preset-vaapi from the main streams?

[hawkeye217]

Hardware acceleration presets do not get applied internally to streams with the record role, so you could just try specifying the appropriate QSV preset.

[KDPate]

What are your detect streams resolutions and fps?

[bigadz87]

Same settings on the ReoLink cameras

[KDPate]

So the two eufy streams are 4608*1728 for detect? That'll add signficantly to the cpu load.

[bigadz87]

No. I have 3 ReoLink cameras (one being exchanged atm so commented out of config) and 2 X eufy indoor 2k cameras. The ReoLinks use the sub stream (fluent) for detection and the clear stream for recording. The eufys are 1080p streams.