Frigate 0.16 Support for Nvidia 1080?

[sbrinkerhoff]

I am trying to setup Frigate for the first time using an nvidia 1080. I have never had a working Frigate configuration before, this is my initial usage.

My configuration loaded and showed the video stream without the detectors or models sections present.

mqtt:
  enabled: false

go2rtc:
  streams:
   main-833a:
      - rtsp://admin:[email protected]:554/h264Preview_01_sub

cameras:
  main_street: # <------ Name the camera
    enabled: true
    ffmpeg:
      inputs:
        - path: rtsp://192.168.40.187:8554/main-833a # <----- The stream you want to use for detection
          roles:
            - detect
detectors:
  onnx:
    type: onnx

model:
  model_type: yolo-generic
  path: /config/model_cache/yolov7-320.onnx
  width: 320
  height: 320
  input_tensor: nchw
  labelmap_path: /labelmap/coco-80.txt
detect:
  enabled: true
version: 0.16-0
ffmpeg:
  hwaccel_args: preset-nvidia

With the yolov7-320 model loaded, Frigate seems to restart over and over. It seems caused by this.

frigate    | 2025-09-07 17:50:54.861829857  [2025-09-07 17:50:54] frigate.api.fastapi_app        INFO    : Starting FastAPI app
frigate    | 2025-09-07 17:50:54.862704501  Process detector:onnx:
frigate    | 2025-09-07 17:50:54.862709442  Traceback (most recent call last):
frigate    | 2025-09-07 17:50:54.862711338    File "/usr/lib/python3.11/multiprocessing/process.py", line 314, in _bootstrap
frigate    | 2025-09-07 17:50:54.862718910      self.run()
frigate    | 2025-09-07 17:50:54.862721802    File "/opt/frigate/frigate/util/process.py", line 41, in run_wrapper
frigate    | 2025-09-07 17:50:54.862723190      return run(*args, **kwargs)
frigate    | 2025-09-07 17:50:54.862746553             ^^^^^^^^^^^^^^^^^^^^
frigate    | 2025-09-07 17:50:54.862748452    File "/usr/lib/python3.11/multiprocessing/process.py", line 108, in run
frigate    | 2025-09-07 17:50:54.862750278      self._target(*self._args, **self._kwargs)
frigate    | 2025-09-07 17:50:54.862751939    File "/opt/frigate/frigate/object_detection/base.py", line 112, in run_detector
frigate    | 2025-09-07 17:50:54.862753722      object_detector = LocalObjectDetector(detector_config=detector_config)
frigate    | 2025-09-07 17:50:54.862755302                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
frigate    | 2025-09-07 17:50:54.862784243    File "/opt/frigate/frigate/object_detection/base.py", line 57, in __init__
frigate    | 2025-09-07 17:50:54.862786019      self.detect_api = create_detector(detector_config)
frigate    | 2025-09-07 17:50:54.862802843                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
frigate    | 2025-09-07 17:50:54.862804745    File "/opt/frigate/frigate/detectors/__init__.py", line 18, in create_detector
frigate    | 2025-09-07 17:50:54.862806207      return api(detector_config)
frigate    | 2025-09-07 17:50:54.862807401             ^^^^^^^^^^^^^^^^^^^^
frigate    | 2025-09-07 17:50:54.862809103    File "/opt/frigate/frigate/detectors/plugins/onnx.py", line 53, in __init__
frigate    | 2025-09-07 17:50:54.862810734      self.model = ort.InferenceSession(
frigate    | 2025-09-07 17:50:54.862812071                   ^^^^^^^^^^^^^^^^^^^^^
frigate    | 2025-09-07 17:50:54.862814070    File "/usr/local/lib/python3.11/dist-packages/onnxruntime/capi/onnxruntime_inference_collection.py", line 465, in __init__
frigate    | 2025-09-07 17:50:54.862833501      self._create_inference_session(providers, provider_options, disabled_optimizers)
frigate    | 2025-09-07 17:50:54.862836424    File "/usr/local/lib/python3.11/dist-packages/onnxruntime/capi/onnxruntime_inference_collection.py", line 526, in _create_inference_session
frigate    | 2025-09-07 17:50:54.862838331      sess = C.InferenceSession(session_options, self._model_path, True, self._read_config_from_model)
frigate    | 2025-09-07 17:50:54.862840112             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
frigate    | 2025-09-07 17:50:54.862845321  onnxruntime.capi.onnxruntime_pybind11_state.Fail: [ONNXRuntimeError] : 1 : FAIL : Load model from /config/model_cache/yolov7-320.onnx failed:/onnxruntime_src/onnxruntime/core/graph/model.cc:180 onnxruntime::Model::Model(onnx::ModelProto&&, const onnxruntime::PathString&, const onnxruntime::IOnnxRuntimeOpSchemaRegistryList*, const onnxruntime::logging::Logger&, const onnxruntime::ModelOptions&) Unsupported model IR version: 12, max supported IR version: 10

Researching this I found no clear solution or description of what the problem here is. I generated the models this way (from the docs)

  162  git clone https://github.com/NateMeyer/tensorrt_demos
  163  cd tensorrt_demos/yolo
  164  ./download_yolo.sh
  165  python3 yolo_to_onnx.py -m yolov7-320
  166  pip install onnx
  167  python3 yolo_to_onnx.py -m yolov7-320
  168  pip3 install "numpy<2"
  169  python3 yolo_to_onnx.py -m yolov7-320

[inkeep-ai-assistant[bot]]

Based on the documentation, Frigate supports NVIDIA GTX 1080 for object detection using the ONNX detector with TensorRT acceleration (1). The GTX 1080 meets the minimum hardware requirements as it supports CUDA Compute Capability 6.1 and is compatible with the 12.x series CUDA libraries (2).

For your NVIDIA GTX 1080 setup, you should use the -tensorrt Frigate image instead of the standard image (1). Here's the correct configuration:

detectors:
  onnx:
    type: onnx

model:
  model_type: yolo-generic
  width: 320
  height: 320
  input_tensor: nchw
  input_dtype: float
  path: /config/model_cache/yolo.onnx
  labelmap_path: /labelmap/coco-80.txt

(1)

The error you're encountering ("Unsupported model IR version: 12, max supported IR version: 10") indicates that your ONNX model was created with a newer version of ONNX that isn't compatible with the ONNX runtime in your Frigate container.

For NVIDIA GPUs, TensorRT will automatically be detected and used with the ONNX detector in the -tensorrt Frigate image (3). You'll need to ensure you're using the correct Docker image: ghcr.io/blakeblackshear/frigate:stable-tensorrt (4).

Your Docker configuration should include proper GPU access:

services:
  frigate:
    image: ghcr.io/blakeblackshear/frigate:stable-tensorrt
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              device_ids: ['0']
              count: 1
              capabilities: [gpu]

(4)

The minimum driver version on your host system must be >=545, and your GPU must support Compute Capability 5.0 or greater (2). The GTX 1080 meets these requirements.

📚 Sources:

1. Supported Hardware
2. Recommended hardware
3. Supported Hardware
4. Video Decoding

---

Have another question? Just tag @inkeep.

[NickM-27]

It is technically supported, I think you'll just need to make adjustments so a compatible version of ONNX is used. I am not entirely sure what steps are needed, but Gemini / ChatGPT may be able to adjust the yolov9 script for you to be compatible

[NickM-27]

also, what driver version do you have installed

[NickM-27]

you should try the latest version of the script, which should now work