[Config Support]: High CPU Load #20089

Pendragon77 · 2025-09-16T09:10:39Z

Pendragon77
Sep 16, 2025

Describe the problem you are having

I ad installed Frigate on my docker for a year already. At the beginning I bought a Coral TPU und had experienced high CPU load due to ffmpeg.
Now I put some more effort into it and got a GPU (NVIDIA P400) to minimize the ffmpeg CPU load. After some struggle the GPU und TPU seems to work in the docker in Frigate.

I still experience CPU loads wich I will not tolerate, of about 60-70% with all (5) cameras. When I reduce the cameras to a minimal config of 2, I get 30-40% load. Here is a graphical overview, after starting frigate:

The processes in question are

frigate.detector.coral
frigate.process:Haustuer

I already tried to start anew with a clean config and only one camera, but even with one camera and also a reduced detect-solution, the CPU load rose considerably.

Version

0.16.1-e664cb2

Frigate config file

# yaml-language-server: $schema=http://192.168.2.179:5000/api/config/schema
detectors:
  coral:
    type: edgetpu
    device: usb
   # num_threads: 2


ffmpeg:
  #hwaccel_args: preset-nvidia-h264
  hwaccel_args: preset-nvidia-h264
mqtt:
  host: 192.168.1.110
  user: user
  password: password

#Include all cameras by default in Birdseye view
birdseye:
  enabled: False

audio:
  # Optional: Enable audio events (default: shown below)
  enabled: False

go2rtc:
  streams:
    haustuer:
      - ffmpeg:rtsp://user:[email protected]:554/live
    haustuer_sub:
      - ffmpeg:rtsp://user:[email protected]:554/cam/realmonitor?channel=1&subtype=1
    #garage:
    #  - ffmpeg:rtsp://user:[email protected]:554/live
    #garage_sub:
    #  - ffmpeg:rtsp://user:[email protected]:554/cam/realmonitor?channel=1&subtype=1
    terasse:
      - ffmpeg:rtsp://user:[email protected]:554/live
    terasse_sub:
      - ffmpeg:rtsp://user:[email protected]:554/cam/realmonitor?channel=1&subtype=1
    #westfassade:
    #  - ffmpeg:rtsp://user:[email protected]:554/live
    #westfassade_sub:
    #  - ffmpeg:rtsp://user:[email protected]:554/cam/realmonitor?channel=1&subtype=1
    #schuppen:
    #  - ffmpeg:rtsp://user:[email protected]:554/live
    #schuppen_sub:
    #  - ffmpeg:rtsp://user:[email protected]:554/cam/realmonitor?channel=1&subtype=1

cameras:
  Haustuer:
    ffmpeg:
      hwaccel_args: preset-nvidia-h264
      inputs:
        - path: rtsp://127.0.0.1:8554/haustuer
          roles:
            - record
        - path: rtsp://127.0.0.1:8554/haustuer_sub  
          roles:
            - detect 
    detect:
      width: 1920
      height: 1080
      fps: 5
      enabled: true
    zones:
      strasse:
        coordinates: 687,226,865,56,1032,40,1207,124,1343,21,936,0,0,333,0,781
      vorgarten:
        coordinates: 363,1080,1755,1028,1623,53,1248,65,1034,241,656,266,257,487
      Strasse:
        coordinates: 0.172,0.406,0.074,0.282,0.301,0.117,0.384,0.213
        loitering_time: 0
        speed_threshold: 10
        distances: 5,9,5,8
        objects:
          - bicycle
          - bus
          - car
          - cat
          - dog
          - horse
          - person
    objects:
      filters:
        car: {}
    snapshots:
      required_zones: [strasse, vorgarten]
    record:
      enabled: true
      alerts:
        retain:
          days: 10
         # mode: all
      detections:
        retain:
          days: 1
    review:
      alerts:
        required_zones: strasse
    motion:
      mask:
        - 0.39,0.392,0.418,0.11,0.524,0.013,0.684,0.012,0.689,0.296
        - 0.237,0.812,0.076,0.312,0.003,0.318,0.005,0.839,0.155,0.985

  Terasse:
    ffmpeg:
      hwaccel_args: preset-nvidia
      inputs:
        - path: rtsp://127.0.0.1:8554/terasse
          roles:
            - record
        - path: rtsp://127.0.0.1:8554/terasse_sub
          roles:
            - detect  
    detect:
      width: 704  # <---- update for your camera's resolution
      height: 576  # <---- update for your camera's resolution
      fps: 5
      enabled: true
    zones:
      Pflaster:
        coordinates: 
          0.161,0.265,0.279,0.196,0.263,0.089,0.389,0.003,0.439,0.023,0.408,0.12,0.927,0.651,0.812,0.994,0.308,0.99
        #loitering_time: 0
      Rasen:
        coordinates: 
          0.923,0.629,0.997,0.338,0.982,0.097,0.839,0.006,0.445,0.008,0.417,0.115
       # loitering_time: 0
    motion:
      mask: 0.788,0.043,0.98,0.302,0.983,0.04
  
  
record:
  retain:
    # Optional: Number of days to retain recordings regardless of events (default: shown below)
    # NOTE: This should be set to 0 and retention should be defined in events section below
    #       if you only want to retain recordings of events.
    days: 2
    # Optional: Event recording settings
  alerts:
    retain:
      days: 5
      mode: motion
    pre_capture: 5
    post_capture: 8
    
  detections:
    retain:
      days: 5
      mode: motion
    pre_capture: 5
    post_capture: 5
      
snapshots:
   # Optional: Enable writing jpg snapshot to /media/frigate/clips (default: shown below)
   # This value can be set via MQTT and will be updated in startup based on retained value
  enabled: true
   # Optional: save a clean PNG copy of the snapshot image (default: shown below)
  clean_copy: true
   # Optional: print a timestamp on the snapshots (default: shown below)
  timestamp: false
   # Optional: draw bounding box on the snapshots (default: shown below)
  bounding_box: true
   # Optional: crop the snapshot (default: shown below)
  crop: false
   # Optional: height to resize the snapshot to (default: original size)
   #height: 175
   # Optional: Restrict snapshots to objects that entered any of the listed zones (default: no required zones)
#   required_zones: [strasse]
   # Optional: Camera override for retention settings (default: global values)
  retain:
     # Required: Default retention days (default: shown below)
    default: 10
     # Optional: Per object retention days
    objects:
      car: 2
      person: 15


detect:
  stationary:
    interval: 5
    threshold: 50

  enabled: true
objects:
  track:
    - person
    - bicycle
    - car
    - motorcycle
    - bus
    - bird
    - cat
    - dog
    - horse
    - frisbee
    - sports ball
    - kite
    - skateboard
    - bottle
    - cell phone

timestamp_style:
  format: '%Y-%m-%d %H:%M:%S'

# environment_vars:
#  PLUS_API_KEY: "xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx"
version: 0.16-0
camera_groups:
  Vorne-Hinten:
    order: 1
    icon: LuAlignCenter
    cameras:
      - Haustuer
      - Terasse
semantic_search:
  enabled: false
  model_size: small
face_recognition:
  enabled: False
  model_size: small
lpr:
  enabled: False
classification:
  bird:
    enabled: false

Relevant Frigate log output

2025-09-16 10:20:09.998945583  [INFO] Preparing Frigate...
2025-09-16 10:20:10.586117960  [INFO] Starting Frigate...
2025-09-16 10:20:15.168475822  [2025-09-16 10:20:15] frigate.util.config            INFO    : Checking if frigate config needs migration...
2025-09-16 10:20:15.211279048  [2025-09-16 10:20:15] frigate.util.config            INFO    : frigate config does not need migration...
2025-09-16 10:20:15.352544648  [2025-09-16 10:20:15] frigate.app                    INFO    : Starting Frigate (0.16.1-e664cb2)
2025-09-16 10:20:15.480142992  [2025-09-16 10:20:15] peewee_migrate.logs            INFO    : Starting migrations
2025-09-16 10:20:15.480608965  [2025-09-16 10:20:15] peewee_migrate.logs            INFO    : There is nothing to migrate
2025-09-16 10:20:15.854252363  [2025-09-16 10:20:15] frigate.app                    INFO    : Recording process started: 388
2025-09-16 10:20:15.854932048  [2025-09-16 10:20:15] frigate.app                    INFO    : Review process started: 390
2025-09-16 10:20:15.858904820  [2025-09-16 10:20:15] frigate.app                    INFO    : go2rtc process pid: 127
2025-09-16 10:20:15.872584238  [2025-09-16 10:20:15] detector.coral                 INFO    : Starting detection process: 413
2025-09-16 10:20:15.873157328  [2025-09-16 10:20:15] frigate.detectors.plugins.edgetpu_tfl INFO    : Attempting to load TPU as usb
2025-09-16 10:20:15.883208137  [2025-09-16 10:20:15] frigate.app                    INFO    : Output process started: 431
2025-09-16 10:20:17.035062897  [2025-09-16 10:20:17] frigate.app                    INFO    : Camera processor started for Haustuer: 458
2025-09-16 10:20:17.042070884  [2025-09-16 10:20:17] frigate.app                    INFO    : Camera processor started for Terasse: 460
2025-09-16 10:20:17.053615608  [2025-09-16 10:20:17] frigate.app                    INFO    : Capture process started for Haustuer: 467
2025-09-16 10:20:17.080489778  [2025-09-16 10:20:17] frigate.app                    INFO    : Capture process started for Terasse: 487
2025-09-16 10:20:17.250711068  [2025-09-16 10:20:17] frigate.api.fastapi_app        INFO    : Starting FastAPI app
2025-09-16 10:20:17.443561536  [2025-09-16 10:20:17] frigate.api.fastapi_app        INFO    : FastAPI started
2025-09-16 10:20:18.530505712  [2025-09-16 10:20:18] frigate.detectors.plugins.edgetpu_tfl INFO    : TPU found
2025-09-16 10:20:18.543524689  INFO: Created TensorFlow Lite XNNPACK delegate for CPU.

Relevant go2rtc log output

2025-09-16 10:20:10.586472194  [INFO] Preparing new go2rtc config...
2025-09-16 10:20:11.028596935  [INFO] Starting go2rtc...
2025-09-16 10:20:11.134661066  10:20:11.134 INF go2rtc platform=linux/amd64 revision=fa580c5 version=1.9.9
2025-09-16 10:20:11.134666111  10:20:11.134 INF config path=/dev/shm/go2rtc.yaml
2025-09-16 10:20:11.135073507  10:20:11.135 INF [api] listen addr=:1984
2025-09-16 10:20:11.135131441  10:20:11.135 INF [rtsp] listen addr=:8554
2025-09-16 10:20:11.135450027  10:20:11.135 INF [webrtc] listen addr=:8555
2025-09-16 10:20:19.732057538  [INFO] Starting go2rtc healthcheck service...

Frigate stats

No response

Operating system

Debian

Install method

Docker Compose

docker-compose file or Docker CLI command

services:
  Frigate:
    container_name: Frigate
    image: ghcr.io/blakeblackshear/frigate:stable
    cap_add:
      - CAP_PERFMON
    privileged: true # this may not be necessary for all setups
    shm_size: "370mb" # update for your cameras based on calculation above
    devices:
      - /dev/bus/usb:/dev/bus/usb
    #  - /dev/dri:/dev/dri
    # GPU Integration
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              #device_ids: ['0'] # this is only needed when using multiple GPUs
              count: 1 # number of GPUs
              capabilities: [gpu]
    runtime: nvidia  
    environment:
      - PLUS_API_KEY=xxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
    hostname: "4340d6d45b9f"
    ipc: "private"
    logging:
      driver: "json-file"
      options: {}
    mac_address: "02:42:ac:11:00:09"
    network_mode: "bridge"
    ports:
      - 1935:1935/tcp
      - 5000:5000/tcp
    restart: "unless-stopped"
    volumes:
      - /srv/dev-disk-by-uuid-9ec9b5a8-c948-4286-b43a-995fb00d8aec/Multimedia/CONFIG/frigate/config:/config
      - /srv/dev-disk-by-uuid-9ec9b5a8-c948-4286-b43a-995fb00d8aec/Multimedia/CONFIG/frigate/media:/media/frigate
      - /etc/localtime:/etc/localtime:ro
      - /usr/lib/x86_64-linux-gnu/libedgetpu.so.1:/usr/lib/x86_64-linux-gnu/libedgetpu.so.1
      - type: tmpfs # Optional: 1GB of memory, reduces SSD/SD Card wear
        target: /tmp/cache
        tmpfs:
          size: 1000000000
    working_dir: "/opt/frigate"