[HW Accel Support]: Multiple Nvidia GPUs

[sohail579]

Describe the problem you are having

I am trying to get frigate to use my GPU I have 2 Nvidia GPUs in my system. I am running on Proxmox LXC with Docker. I have managed to install the required drivers all is working except its using the wrong GPU. its using the card 0 when I have set it to sue card 1 in my compose file

Version

0.13.2-6476F8A

Frigate config file

mqtt:
  enabled: false

detectors:
 coral:
   type: edgetpu
   device: usb

ffmpeg:
  hwaccel_args:
    - -c:v
    - h264_cuvid

go2rtc:
  streams:
    Front_Garden:
      - rtsp://admin:[email protected]:554/Streaming/Channels/301
    Front_Garden_sub:
      - rtsp://admin:[email protected]:554/Streaming/Channels/302

cameras:
  Front_Garden:
    ffmpeg:
      inputs:
        - path: rtsp://127.0.0.1:8554/Front_Garden_sub
          roles:
            - detect
        - path: rtsp://127.0.0.1:8554/Front_Garden?video=copy&audio=aac
          roles:
            - record

    detect:
      width: 640
      height: 480
      fps: 25

    snapshots:
      enabled: True

    record:
      enabled: True
      retain:
        days: 0
        mode: motion
      events:
        retain:
          default: 5
          mode: active_objects
        pre_capture: 5
        post_capture: 8
        required_zones:
          - Front_Path
    motion:
      mask:
        - 0,51,242,47,255,0,0,0
        - 80,480,640,480,640,268,299,122,38,258

    zones:
      Front_Path:
        coordinates: 29,311,235,164,340,168,441,187,491,155,423,120,355,90,325,85,286,71,214,72,162,40,21,70
        objects:
          - person
        filters:
          person:
            min_area: 5000
            max_area: 100000
           # threshold: 0.7

docker-compose file or Docker CLI command

version: "3.9"
services:
  frigate:
    container_name: frigate
    privileged: true # this may not be necessary for all setups
    restart: unless-stopped
    image: ghcr.io/blakeblackshear/frigate:stable
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              device_ids:
                - "1"
              capabilities:
                - gpu
    shm_size: 64mb # update for your cameras based on calculation above
    devices:
      - /dev/bus/usb:/dev/bus/usb
    volumes:
      - /etc/localtime:/etc/localtime:ro
      - /frigate/config:/config
      - /recordings:/media/frigate
      - type: tmpfs
        target: /tmp/cache
        tmpfs:
          size: 1000000000
    ports:
      - 5000:5000
      - 8554:8554 # RTSP feeds
      - 8555:8555/tcp # WebRTC over tcp
      - 8555:8555/udp # WebRTC over udp
    environment:
      FRIGATE_RTSP_PASSWORD: password
networks: {}

Relevant log output

2024-07-17 18:23:07.988939422  [INFO] Preparing Frigate...
2024-07-17 18:23:08.001495088  [INFO] Starting Frigate...
2024-07-17 18:23:08.953042251  [2024-07-17 18:23:08] frigate.app                    INFO    : Starting Frigate (0.13.2-6476f8a)
2024-07-17 18:23:08.979669084  [2024-07-17 18:23:08] peewee_migrate.logs            INFO    : Starting migrations
2024-07-17 18:23:08.987697015  [2024-07-17 18:23:08] peewee_migrate.logs            INFO    : There is nothing to migrate
2024-07-17 18:23:08.987814897  [2024-07-17 18:23:08] frigate.app                    INFO    : Recording process started: 2279
2024-07-17 18:23:08.990152629  [2024-07-17 18:23:08] frigate.app                    INFO    : go2rtc process pid: 89
2024-07-17 18:23:09.012700674  [2024-07-17 18:23:09] frigate.app                    INFO    : Output process started: 2290
2024-07-17 18:23:09.029827177  [2024-07-17 18:23:09] frigate.app                    INFO    : Camera processor started for Front_Garden: 2296
2024-07-17 18:23:09.030764987  [2024-07-17 18:23:09] frigate.app                    INFO    : Capture process started for Front_Garden: 2298
2024-07-17 18:23:11.602811363  [2024-07-17 18:23:09] detector.coral                 INFO    : Starting detection process: 2288
2024-07-17 18:23:11.602859284  [2024-07-17 18:23:09] frigate.detectors.plugins.edgetpu_tfl INFO    : Attempting to load TPU as usb
2024-07-17 18:23:11.606323241  [2024-07-17 18:23:11] frigate.detectors.plugins.edgetpu_tfl INFO    : TPU found

FFprobe output from your camera

n/a

Operating system

Proxmox

Install method

Docker Compose

Network connection

Wired

Camera make and model

Hikvision

Any other information that may be helpful

No response

[NickM-27]

I see a few issues in the config

1. should be using preset-nvidia-h264 instead of manual args for hwaccel_args
2. detect fps should not be that high, it does not need to match camera fps

as far as this issue, remove privileged from the container

[sohail579]

You, sir, are a genius. Worked a treat. Some questions if possible. 1. So that I understand what was happening there? 2. Will go2rtc use the gpu too? 3. My coral inference has jumped from 7ms to 20ms is that an issue?

[NickM-27]

1. When a preset is used, this tells frigate to figure out the best args that are more efficient than using manual args.
2. In most cases no, because there is nothing for the GPU to do unless transcoding is occurring.
3. Seems odd, no idea why that would happen. Not a huge issue as long as frames aren't being skipped