[HW Accel Support]:

[gcardy]

Describe the problem you are having

I have configured hardware acceleration, but I am seeing CPU utilization and next to no GPU utilization on my system. I thought that hardware acceleration would offload the ffmpeg load onto the GPU. Am I misunderstanding something?

Version

0.13.2-6476f8a

Frigate config file

mqtt:
  host: mosquitto
  user: mqttuser
  password: xxxxxxx

detectors:
  coral1:
    type: edgetpu
    device: pci:0

record:
  enabled: true
  retain:
    days: 7
  events:
    retain:
      default: 45

snapshots:
  enabled: true
  timestamp: true
  bounding_box: true

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

ffmpeg:
  hwaccel_args: preset-nvidia-h264

go2rtc:
  streams:
    driveway_high_res:
    - rtsp://xxx:[email protected]/cam/realmonitor?channel=1&subtype=0&authbasic=64

cameras:
  driveway:
    ffmpeg:
      inputs:
      - path: rtsp://localhost:8554/driveway_high_res
        roles:
        - record
        - detect

docker-compose file or Docker CLI command

version: "3.9"
services:
  frigate.geonkins.net:
    container_name: frigate
    privileged: true # this may not be necessary for all setups
    restart: unless-stopped
    image: ghcr.io/blakeblackshear/frigate:stable-tensorrt
    runtime: nvidia
    shm_size: "750mb"
    networks:
      net:
        ipv4_address: 192.168.0.195
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu]
    devices:
      - /dev/apex_0:/dev/apex_0
    volumes:
      - '/etc/localtime:/etc/localtime:ro'
      - '/vol01/frigate/storage:/media/frigate'
      - '/vol01/frigate/config:/config'
      - type: tmpfs # Optional: 1GB of memory, reduces SSD/SD Card wear
        target: /tmp/cache
        tmpfs:
          size: 4000000000
    environment:
      NVIDIA_VISIBLE_DEVICES: all
      NVIDIA_DRIVER_CAPABILITIES: "compute,utility,video"
      FRIGATE_RTSP_PASSWORD: "xxxxxxxxxxxx"
networks:
  net:
    name: net
    external: true

Relevant log output

n/a

FFprobe output from your camera

version 5.1.4-0+deb12u1 Copyright (c) 2007-2023 the FFmpeg developers
  built with gcc 12 (Debian 12.2.0-14)
  configuration: --prefix=/usr --extra-version=0+deb12u1 --toolchain=hardened --libdir=/usr/lib/x86_64-linux-gnu --incdir=/usr/include/x86_64-linux-gnu --arch=amd64 --enable-gpl --disable-stripping --enable-gnutls --enable-ladspa --enable-libaom --enable-libass --enable-libbluray --enable-libbs2b --enable-libcaca --enable-libcdio --enable-libcodec2 --enable-libdav1d --enable-libflite --enable-libfontconfig --enable-libfreetype --enable-libfribidi --enable-libglslang --enable-libgme --enable-libgsm --enable-libjack --enable-libmp3lame --enable-libmysofa --enable-libopenjpeg --enable-libopenmpt --enable-libopus --enable-libpulse --enable-librabbitmq --enable-librist --enable-librubberband --enable-libshine --enable-libsnappy --enable-libsoxr --enable-libspeex --enable-libsrt --enable-libssh --enable-libsvtav1 --enable-libtheora --enable-libtwolame --enable-libvidstab --enable-libvorbis --enable-libvpx --enable-libwebp --enable-libx265 --enable-libxml2 --enable-libxvid --enable-libzimg --enable-libzmq --enable-libzvbi --enable-lv2 --enable-omx --enable-openal --enable-opencl --enable-opengl --enable-sdl2 --disable-sndio --enable-libjxl --enable-pocketsphinx --enable-librsvg --enable-libmfx --enable-libdc1394 --enable-libdrm --enable-libiec61883 --enable-chromaprint --enable-frei0r --enable-libx264 --enable-libplacebo --enable-librav1e --enable-shared
  libavutil      57. 28.100 / 57. 28.100
  libavcodec     59. 37.100 / 59. 37.100
  libavformat    59. 27.100 / 59. 27.100
  libavdevice    59.  7.100 / 59.  7.100
  libavfilter     8. 44.100 /  8. 44.100
  libswscale      6.  7.100 /  6.  7.100
  libswresample   4.  7.100 /  4.  7.100
  libpostproc    56.  6.100 / 56.  6.100
[rtsp @ 0x55ed7a150700] max delay reached. need to consume packet
[rtsp @ 0x55ed7a150700] RTP: missed 35 packets
Input #0, rtsp, from 'rtsp://x:[email protected]':
  Metadata:
    title           : Media Server
  Duration: N/A, start: 0.000000, bitrate: N/A
  Stream #0:0: Video: h264 (Main), yuv420p(progressive), 3840x2160, 30 fps, 100 tbr, 90k tbn
  Stream #0:1: Audio: aac (LC), 64000 Hz, mono, fltp

### Operating system

Debian

### Install method

Docker Compose

### Network connection

Wired

### Camera make and model

Amcrest IP8M-T2599EW 

### Any other information that may be helpful

I know that I have detection setup on a high-res stream, and that is resource intensive, but I would expect ffmpeg to use some GPU resources and not CPU exclusively. ```

[NickM-27]

It would be best to see a screenshot of the system page from the frigate webUI

[gcardy]

Here you go, appreciate the quick response.

[NickM-27]

yeah, these numbers look reasonable for decoding 4k. The GPU can only do so much, the CPU still needs to move the data to GPU memory and back as well as convert the pixel format. I imagine you would see a considerable decrease if you manually set the detect resolution to 1920x1080 which would use the GPU to do the scaling

[gcardy]

Thanks Nick! Setting the detect resolution has helped significantly. With all 10 of my cameras being ingested I'm seeing much better performance.