[HW Accel Support]: NVidia GPU 0%

[anjoze]

Describe the problem you are having

I have a HP Proliant G6 with a Nvidia Quadro P620
Truenas Ver. 26.0.0-BETA.1

I’ve setup Frigate and it work very well but the CPU is too high so, I added the Nvidia Quadro to process the detection.

The Quadro P620 insn’t supported anymore by the latest driver so I had to follow this instructions:
https://github.com/zzzhouuu/truenas-nvidia-drivers

Everything works as expected except the GPU usage, its always 0%.

Because this is installed as a APP the GPU selection is in the App settings:

Then I search here for the GPU info:

/mnt/.ix-apps/app_configs/frigate/metadata.yaml
/mnt/.ix-apps/app_configs/[app]/versions/1.3.8/user_config.yaml

The APP is running with root permissions:

Version

0.17.1-416a9b7

Frigate config file

mqtt:
  enabled: false

# Keep your hardware accelerated decoding configuration
ffmpeg:
  hwaccel_args: preset-nvidia-h264
  output_args:
    record: preset-record-generic-audio-aac

go2rtc:
  streams:
    cam_f55bed20:
      - rtsp://{FRIGATE_RTSP_USER}:{FRIGATE_RTSP_PASSWORD}@192.168.1.201:554/stream1
    cam_f55bed20_sub:
      - rtsp://{FRIGATE_RTSP_USER}:{FRIGATE_RTSP_PASSWORD}@192.168.1.201:554/stream2

detectors:
  onnx_0:
    type: onnx
  onnx_1:
    type: onnx

model:
  model_type: yolo-generic
  width: 320 # <--- should match the imgsize set during model export
  height: 320 # <--- should match the imgsize set during model export
  input_tensor: nchw
  input_dtype: float
  #input_pixel_format: rgb
  path: /config/model_cache/yolov9-t-320.onnx
  #labelmap_path: /config/labelmap/coco-80.txt

detect:
  enabled: true
  width: 640
  height: 360
  fps: 5

objects:
      track:
        - person
        - car

record:
  sync_recordings: true # Sync your DB and records
  enabled: true
  motion:
    days: 1
  alerts:
    retain:
      days: 30
      mode: motion
  detections:
    retain:
      days: 30
      mode: motion

cameras:     # No cameras defined, UI wizard should be used
  cam_f55bed20:
    enabled: true
    friendly_name: Main Gate
    ffmpeg:
      inputs:
        - path: rtsp://127.0.0.1:8554/cam_f55bed20
          roles:
            - record
        - path: rtsp://127.0.0.1:8554/cam_f55bed20_sub
          roles:
            - detect

    live:
      streams:
        Stream 1: cam_f55bed20

    zones:
      zona1:
        coordinates: 0,0.29,1,0.123,1,1,0.001,0.996
        loitering_time: 0
        friendly_name: Zona1

    motion:
      mask: 
        0.175,0.095,0.334,0.035,0.401,0.207,0.362,0.356,0.284,0.399,0.205,0.304
    notifications:
      enabled: true

    review:
      detections:
        labels:
          - car
          - cat
          - dog
          - person
        required_zones: zona1
      alerts:
        labels:
          - car
          - cat
          - dog
          - person
        required_zones: zona1
version: 0.17-0

docker-compose file or Docker CLI command

Generated from Truenas APP

Relevant Frigate log output

2026-06-01 10:57:10.479790+01:002026-06-01 09:57:10.479649899  [2026-06-01 10:57:10] frigate.util.config            INFO    : Checking if frigate config needs migration...
2026-06-01 10:57:10.515787+01:002026-06-01 09:57:10.515574772  [2026-06-01 10:57:10] frigate.util.config            INFO    : frigate config does not need migration...
2026-06-01 10:57:10.612973+01:002026-06-01 09:57:10.612786486  [2026-06-01 10:57:10] frigate.app                    INFO    : Starting Frigate (0.17.1-416a9b7)
2026-06-01 10:57:10.678993+01:002026-06-01 09:57:10.678687104  [2026-06-01 10:57:10] peewee_migrate.logs            INFO    : Starting migrations
2026-06-01 10:57:10.680480+01:002026-06-01 09:57:10.680363828  [2026-06-01 10:57:10] peewee_migrate.logs            INFO    : There is nothing to migrate
2026-06-01 10:57:10.766634+01:002026-06-01 09:57:10.766456437  [2026-06-01 10:57:10] frigate.app                    INFO    : Recording process started: 528
2026-06-01 10:57:10.792712+01:002026-06-01 09:57:10.792434009  [2026-06-01 10:57:10] frigate.app                    INFO    : Review process started: 531
2026-06-01 10:57:10.798272+01:002026-06-01 09:57:10.798094504  [2026-06-01 10:57:10] frigate.app                    INFO    : go2rtc process pid: 150
2026-06-01 10:57:11.432946+01:002026-06-01 09:57:11.432672651  [2026-06-01 10:57:11] frigate.app                    INFO    : Embedding process started: 551
2026-06-01 10:57:11.470300+01:002026-06-01 09:57:11.470067826  [2026-06-01 10:57:11] frigate.detectors.plugins.onnx INFO    : ONNX: loading /config/model_cache/yolov9-t-320.onnx
2026-06-01 10:57:11.490999+01:002026-06-01 09:57:11.490778458  [2026-06-01 10:57:11] frigate.detectors.plugins.onnx INFO    : ONNX: loading /config/model_cache/yolov9-t-320.onnx
2026-06-01 10:57:11.540500+01:002026-06-01 09:57:11.540280651  [2026-06-01 10:57:11] frigate.app                    INFO    : Output process started: 612
2026-06-01 10:57:11.614318+01:002026-06-01 09:57:11.614133819  [2026-06-01 10:57:11] frigate.camera.maintainer      INFO    : Camera processor started for cam_f55bed20: 651

Relevant go2rtc log output

info | 2026-06-01 09:56:59 | startup | Preparing new go2rtc config...
info | 2026-06-01 09:57:00 | startup | Starting go2rtc...
info | 2026-06-01 09:57:01 | startup | go2rtc platform=linux/amd64 revision=df95ce3 version=1.9.10
info | 2026-06-01 09:57:01 | startup | config path=/config/go2rtc_homekit.yml
info | 2026-06-01 09:57:01 | rtsp    | listen addr=:8554
info | 2026-06-01 09:57:01 | api     | listen addr=:1984
info | 2026-06-01 09:57:01 | webrtc  | listen addr=:8555
info | 2026-06-01 09:57:09 | startup | Starting go2rtc healthcheck service...

FFprobe output from your camera

Not possible from Trunas

Install method

Docker Compose

Object Detector

Other

Network connection

Wired

Camera make and model

Tapo

Screenshots of the Frigate UI's System metrics pages

Any other information that may be helpful

I've started a discussion on Truenas forum with all the tests I made:
https://forums.truenas.com/t/frigate-nvidia-gpu-0-with-yolov9-and-onnx-detector/66142

[inkeep-ai-assistant[bot]]

Your setup has two separate GPU workloads to consider: video decoding and object detection. Based on your screenshots and config, your video decoding via NVIDIA is actually working—the nvidia-smi output shows /usr/lib/ffmpeg/7.0/bin/ffmpeg using 92MiB of GPU memory, which indicates hardware-accelerated decoding is active. (1)

However, your ONNX detectors are not using the GPU for object detection. To use NVIDIA GPUs for ONNX-based object detection, you must use the Frigate Docker image with the -tensorrt suffix, for example ghcr.io/blakeblackshear/frigate:stable-tensorrt. (2)

The standard Frigate image does not include TensorRT support, so even though you've configured type: onnx detectors, they're running on CPU rather than GPU. This explains why your GPU utilization shows 0% for compute tasks while ffmpeg decoding is working.

To enable GPU-accelerated object detection:

1. Switch to the TensorRT image variant (ghcr.io/blakeblackshear/frigate:stable-tensorrt) (2)
2. Ensure the NVIDIA Container Toolkit is properly configured and the GPU is passed to the container with the appropriate capabilities (1)

Your current config for video decoding (preset-nvidia-h264) is correct and appears to be functioning. (1)

Note that the Quadro P620 should support CUDA/NVDEC—you can verify compatibility on NVIDIA's GPU support matrix. (1)

📚 Sources:

1. Video Decoding
2. Supported Hardware

---

Have another question? Just tag @inkeep.

[anjoze]

Thanks Mr Bot.
Created a new yaml file with
ghcr.io/blakeblackshear/frigate:stable-tensorrt
It works with GPU