Enabled Nvidia-based hardware acceleration but CPU usage does not seem to change

[cmsimike]

Hi,

I've gone through https://docs.frigate.video/configuration/hardware_acceleration/ to setup hardware accelerated video decoding and I believe it's setup as correct as I know how. I confirmed I selected the correct decoder since I can see video in the UI but even changing to different decoders showed a green video stream. For reference here is my ffmpeg config:

      hwaccel_args: -c:v h264_cuvid
      output_args:
        record: -f segment -segment_time 10 -segment_format mp4 -reset_timestamps 1 -strftime 1 -c:v copy -ar 44100 -c:a aac
        rtmp: -c:v copy -f flv -ar 44100 -c:a aac

I tried running docker exec -it frigate nvidia-smi but the output seemed empty?

+-----------------------------------------------------------------------------+
| NVIDIA-SMI 470.141.03   Driver Version: 470.141.03   CUDA Version: 11.4     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|                               |                      |               MIG M. |
|===============================+======================+======================|
|   0  NVIDIA GeForce ...  On   | 00000000:01:00.0 Off |                  N/A |
| 21%   48C    P2    45W / 200W |   1131MiB /  8117MiB |      3%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes:                                                                  |
|  GPU   GI   CI        PID   Type   Process name                  GPU Memory |
|        ID   ID                                                   Usage      |
|=============================================================================|
+-----------------------------------------------------------------------------+

but if I run nvidia-smi on the host machine directly, I am seeing ffmpeg

Sat Oct 22 16:23:52 2022
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 470.141.03   Driver Version: 470.141.03   CUDA Version: 11.4     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|                               |                      |               MIG M. |
|===============================+======================+======================|
|   0  NVIDIA GeForce ...  On   | 00000000:01:00.0 Off |                  N/A |
| 21%   48C    P2    45W / 200W |   1131MiB /  8117MiB |      3%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes:                                                                  |
|  GPU   GI   CI        PID   Type   Process name                  GPU Memory |
|        ID   ID                                                   Usage      |
|=============================================================================|
|    0   N/A  N/A      3062      C   ffmpeg                            161MiB |
|    0   N/A  N/A      3067      C   ffmpeg                            161MiB |
|    0   N/A  N/A      3075      C   ffmpeg                            161MiB |
|    0   N/A  N/A      3081      C   ffmpeg                            161MiB |
|    0   N/A  N/A      3084      C   ffmpeg                            161MiB |
|    0   N/A  N/A      3098      C   ffmpeg                            161MiB |
|    0   N/A  N/A      3102      C   ffmpeg                            161MiB |
+-----------------------------------------------------------------------------+

It worries me I am not seeing the ffmpeg in the container but not sure if that's correct or not. It does diverge from the docs. But as mentioned above, I am seeing the ffmpeg processes on the host.

In any event, my CPU usage has not seemed to change at all. I (maybe incorrectly) would assume that cpu usage would go down or I'd basically see no cpu usage but I am seeing high cpu usage on the host machine from ffmpeg:

I'm just looking to confirm whether or not this is correct and if it isn't how can I go ahead and fix this.

Thanks!

[blakeblackshear]

If you see it from the host it's working. The process in the container is likely missing some permissions.

[cmsimike]

@blakeblackshear Thanks! Any tips on what specific permissions to look at for? I've basically followed this https://docs.frigate.video/configuration/hardware_acceleration/ and installed the required deps (nvidia-cuda-toolkit, nvidia-driver).

[blakeblackshear]

I'm not sure how to get the output to be correct inside the container.

[NickM-27]

There is an issue documented here with the ways around this: NVIDIA/nvidia-docker#179 (comment)

[cmsimike]

Thanks! I am already doing running with privileged: true which seems to be one of the solutions listed in there.