Issues running nvidia acceleration with 0.11

[thinkloop]

I am trying to use an nvidia gpu for hardware acceleration on the 0.11 build using these instructions.

The first issue I am running into is that the instructions say to use the image blakeblackshear/frigate:stable, however I get an error that the manifest for it does not exist: ERROR: manifest for blakeblackshear/frigate:stable not found: manifest unknown: manifest unknown

So I changed it to blakeblackshear/frigate:0.11.0-beta4 and that runs (not sure if I should do this tho).

Next the instructions say to add a "deploy" section:

  deploy:    # <------------- Add this section
    resources:
      reservations:
        devices:
          - driver: nvidia
            count: 1
            capabilities: [gpu]

But the instructions show that section at the same indent level as "frigate" which caused errors, so I indented it one level in so that it is within/under the "frigate" config and that seems to work (not sure if that's correct tho).

Finally I updated the ffmpeg config in config.yml:

ffmpeg:

  # nvidia acceleration
  hwaccel_args: -c:v h264_cuvid

  # enable audio
  output_args:
    record: -f segment -segment_time 10 -segment_format mp4 -reset_timestamps 1 -strftime 1 -c:v copy -c:a aac

Everything runs now, however when I go to "debug" in the ui it says I am still running on CPU:

Any ideas of what I'm doing wrong?

[NickM-27]

The first issue I am running into is that the instructions say to use the image blakeblackshear/frigate:stable .. So I changed it to blakeblackshear/frigate:0.11.0-beta4

These docs are written for the final release, it needs to be the beta image at this time (so we don't have to update the docs for every beta release) so you are correct in using blakeblackshear/frigate:0.11.0-beta4

so I indented it one level in so that it is within/under the "frigate" config and that seems to work (not sure if that's correct tho).

Should be good with the way you indented it, may need to update those docs

Everything runs now, however when I go to "debug" in the ui it says I am still running on CPU:

That is the detector IE the device that is running object detection, not what is doing hardware acceleration. As the docs you linked suggest, you need to use nvidia smi command to ensure there are ffmpeg processes utilizing the GPU

[thinkloop]

Well then we have great success!