Hardware Acc w/ nVidia = No running processes found?

[mlg9000]

I have an old Geforce GT 730 running the 470 driver that from that I can tell supports decoding based on "ffmpeg -decoders | grep cuvid" showing h264_cuvid (this card does not show up on the support matrix for decoders). From what I can also tell I've managed to get it working under docker as I get the same output there. However, running nvidia-smi either in the container or on the host directly I get "No running processes found". There are no errors in the container logs but the CPU usage is high still so it doesn't appear hardware decoding is working. Anyone have any ideas what the issue might be? Do I need to buy a new card?

[NickM-27]

What version of frigate are you running (need the number, not latest)

[mlg9000]

No usage on the GPU. The CPU usage is still as high as it previously was before enabling GPU support.

[NickM-27]

Can you share your config?

[mlg9000]

Can you share your config?

mqtt:
  host: hass.my.domain
  topic_prefix: frigate
  client_id: frigate
  user: frigate
  password: fakepass
cameras:
  shop_front:
    ffmpeg:
      hwaccel_args: -c:v h264_cuvid
      input_args:
        - -avoid_negative_ts
        - make_zero
        - -fflags
        - +genpts+discardcorrupt
        - -flags
        - low_delay
        - -strict
        - experimental
        - -analyzeduration
        - 1000M
        - -probesize
        - 1000M
        - -rw_timeout
        - "5000000"
      inputs:
        - path: rtmp://cam-1.my.domain/bcs/channel0_main.bcs?token=254&channel=0&stream=0&user=user&password=fakepass
          roles:
            - detect
            - rtmp
            - clips
            - record
    detect:
      width: 2560
      height: 1920
      fps: 5
    motion:
      mask:
        - 0,1182,105,934,172,808,266,705,340,687,419,665,483,632,514,592,572,552,622,509,673,434,792,363,835,321,851,210,853,108,904,100,1014,90,1116,81,1196,72,1437,67,1810,89,1994,116,2299,198,2496,0,0,0
        - 997,1920,410,1920,0,1920,0,1412,418,1675
    objects:
      track:
        - person
        - car
      filters:
        car:
          mask:
            - 1719,1920,959,1920,0,1920,0,1505,0,1123,2071,1184,2053,981,2560,995,2560,1920
            - 2154,291,2126,231,2015,226,1874,219,1748,203,1651,199,1512,182,1377,156,1196,95,1098,0,2560,0,2560,848
            - 505,1102,0,1049,0,120,754,123,869,119,949,152,956,289,942,137,981,209,890,387
        person:
          min_area: 5000
          max_area: 100000
          min_score: 0.5
          threshold: 0.72
detectors:
  coral:
    type: edgetpu
    device: usb

[NickM-27]

a few things:

1. clips is no longer a valid role so recommend removing that.
2. on 0.11 we would expect ffmpeg to crash if it can't use hwaccel so that is weird. does it work if you run ffmpeg with -c:v h264_cuvid on your host?

[kylepyke]

I'm having the same issue on a Geforce GT 750m. No running processes detected, no change in CPU. nvidia-smi brings up the same output in docker, and in the host.

[mccasian]

Hi there. Did you managed to find a solution to this?