Hardware acceleration?

[danuals]

Thanks for this, running in a dd220+ with coral usb.

is hardware acceleration possible or a benefit?

[weltenwort]

The Coral USB device is a hardware accelerator for inference. Are you asking about video encoding/decoding acceleration in addition to that?

[danuals]

Yes. Hw accel args, wondered if it’s possible and what benefit it would be

[weltenwort]

They are no different from the official image (https://docs.frigate.video/configuration/hardware_acceleration). I'm using the following config on my intel-based disk station, but it looks like the CPU is still heavily involved in the image scaling:

ffmpeg:
  hwaccel_args:
    - "-hwaccel"
    - "qsv"
    - "-qsv_device"
    - "/dev/dri/renderD128"

So I'm still looking for the right way to configure ffmpeg to use acceleration for that too 🤷 If you find out anything I'd be interested to hear about it.

[danuals]

Ok thanks, will let you know how I go.

[danuals]

so it seems my ds220+ is a 9th gen intel so used the standard args, cpu usage was fairy constant 30-40% in docker stats before, now it drops down to 18 and ramps up to 40 max in waves,

ffmpeg: hwaccel_args: - -hwaccel - vaapi - -hwaccel_device - /dev/dri/renderD128 - -hwaccel_output_format - yuv420p

[weltenwort]

interesting, so it seems the inference produced quite a significant part of the cpu load in your case 👍 for me I suspect the scaling I have in my ffmpeg config causes the problem. the camera doesn't provide a suitably scaled stream. I tried using scale_qsv and scale_vaapi to accelerate it, but no luck so far.