compute graph nodes have not been partitioned to the CUDAExecutionProvider

[Pasha-Man]

Describe the problem you are having

the cameras stream works fine. but the AI detetction does not work. i tried the CPu one, the onvx one, the tensor one. but all show error. i followed the guide but still stuck with it. and it continiously restarts so i cant take ss of the metrcis as well

Version

0.17.1-416a9b7

Frigate config file

mqtt:
  host: mqtt

auth:
  enabled: false


detectors:
  yolov9_detector:
    type: onnx
    device: cuda
    # Remove the environment_vars block from here or the root level
    onnx:
      cuda_capture_enable: false  # This is the correct 0.17+ syntax

model:
  path: /config/model_cache/yolov9-s.onnx
  model_type: yolo-generic
  input_tensor: nchw
  input_pixel_format: rgb
  input_dtype: float
  width: 640
  height: 640

ffmpeg:
  hwaccel_args: preset-nvidia-h264

record:
  enabled: true
  alerts:
    retain:
      days: 4
      mode: motion
  detections:
    retain:
      days: 4
      mode: motion

go2rtc:
  streams:
    camera_1: rtsp://admin:hik.12345@192.168.1.14:554/Streaming/Channels/101
    camera_1_sub: rtsp://admin:hik.12345@192.168.1.14:554/Streaming/Channels/102
    camera_2: rtsp://admin:hik.12345@192.168.1.14:554/Streaming/Channels/201
    camera_2_sub: rtsp://admin:hik.12345@192.168.1.14:554/Streaming/Channels/202
    camera_3: rtsp://admin:hik.12345@192.168.1.14:554/Streaming/Channels/301
    camera_3_sub: rtsp://admin:hik.12345@192.168.1.14:554/Streaming/Channels/302
    camera_4: rtsp://admin:hik.12345@192.168.1.14:554/Streaming/Channels/401
    camera_4_sub: rtsp://admin:hik.12345@192.168.1.14:554/Streaming/Channels/402
    camera_5: rtsp://admin:hik.12345@192.168.1.14:554/Streaming/Channels/501
    camera_5_sub: rtsp://admin:hik.12345@192.168.1.14:554/Streaming/Channels/502
    camera_6: rtsp://admin:hik.12345@192.168.1.14:554/Streaming/Channels/601
    camera_6_sub: rtsp://admin:hik.12345@192.168.1.14:554/Streaming/Channels/602
    camera_7: rtsp://admin:hik.12345@192.168.1.14:554/Streaming/Channels/701
    camera_7_sub: rtsp://admin:hik.12345@192.168.1.14:554/Streaming/Channels/702
    camera_8: rtsp://admin:hik.12345@192.168.1.14:554/Streaming/Channels/801
    camera_8_sub: rtsp://admin:hik.12345@192.168.1.14:554/Streaming/Channels/802

cameras:
  camera_1:
    ffmpeg:
      inputs:
        - path: rtsp://127.0.0.1:8554/camera_1_sub
          roles: [detect]
        - path: rtsp://127.0.0.1:8554/camera_1
          roles: [record]
    detect:
      width: 640
      height: 480
      fps: 5
  camera_2:
    ffmpeg:
      inputs:
        - path: rtsp://127.0.0.1:8554/camera_2_sub
          roles: [detect]
        - path: rtsp://127.0.0.1:8554/camera_2
          roles: [record]
    detect:
      width: 640
      height: 480
      fps: 5
  camera_3:
    ffmpeg:
      inputs:
        - path: rtsp://127.0.0.1:8554/camera_3_sub
          roles: [detect]
        - path: rtsp://127.0.0.1:8554/camera_3
          roles: [record]
    detect:
      width: 640
      height: 480
      fps: 5
  camera_4:
    ffmpeg:
      inputs:
        - path: rtsp://127.0.0.1:8554/camera_4_sub
          roles: [detect]
        - path: rtsp://127.0.0.1:8554/camera_4
          roles: [record]
    detect:
      width: 640
      height: 480
      fps: 5
  camera_5:
    ffmpeg:
      inputs:
        - path: rtsp://127.0.0.1:8554/camera_5_sub
          roles: [detect]
        - path: rtsp://127.0.0.1:8554/camera_5
          roles: [record]
    detect:
      width: 640
      height: 480
      fps: 5
  camera_6:
    ffmpeg:
      inputs:
        - path: rtsp://127.0.0.1:8554/camera_6_sub
          roles: [detect]
        - path: rtsp://127.0.0.1:8554/camera_6
          roles: [record]
    detect:
      width: 640
      height: 480
      fps: 5
  camera_7:
    ffmpeg:
      inputs:
        - path: rtsp://127.0.0.1:8554/camera_7_sub
          roles: [detect]
        - path: rtsp://127.0.0.1:8554/camera_7
          roles: [record]
    detect:
      width: 640
      height: 480
      fps: 5
  camera_8:
    ffmpeg:
      inputs:
        - path: rtsp://127.0.0.1:8554/camera_8_sub
          roles: [detect]
        - path: rtsp://127.0.0.1:8554/camera_8
          roles: [record]
    detect:
      width: 640
      height: 480
      fps: 5
version: 0.17-0

docker-compose file or Docker CLI command

services:
  frigate:
    container_name: frigate
    privileged: true
    restart: unless-stopped
    image: ghcr.io/blakeblackshear/frigate:stable-tensorrt
    shm_size: "2gb"
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu]
    ports:
      - "5050:5000"
      - "8554:8554"
      - "8555:8555/tcp"
      - "8555:8555/udp"
    volumes:
      - /etc/localtime:/etc/localtime:ro
      - ./config:/config
      - ./storage:/media/frigate
      - D:\FrigateLab\models:/models  # <--- THIS IS THE MISSING BRIDGE
      - type: tmpfs 
        target: /tmp/cache
        tmpfs:
          size: 1000000000
  mqtt:
    container_name: mosquitto
    image: eclipse-mosquitto:2
    restart: unless-stopped
    ports:
      - "1883:1883"
    volumes:
      - ./mosquitto/mosquitto.conf:/mosquitto/config/mosquitto.conf

Relevant Frigate log output

2026-06-23 20:27:26.384194356  [INFO] Preparing Frigate...
2026-06-23 20:27:27.155358986  [INFO] Starting Frigate...
2026-06-23 20:27:34.950830467  [2026-06-23 20:27:34] frigate.util.config            INFO    : Checking if frigate config needs migration...
2026-06-23 20:27:34.996446516  [2026-06-23 20:27:34] frigate.util.config            INFO    : frigate config does not need migration...
2026-06-23 20:27:35.360832212  [2026-06-23 20:27:35] frigate.app                    INFO    : Starting Frigate (0.17.1-416a9b7)
2026-06-23 20:27:35.942383959  [2026-06-23 20:27:35] peewee_migrate.logs            INFO    : Starting migrations
2026-06-23 20:27:35.946135776  [2026-06-23 20:27:35] peewee_migrate.logs            INFO    : There is nothing to migrate
2026-06-23 20:27:36.008977787  [2026-06-23 20:27:35] frigate.app                    INFO    : Recording process started: 863
2026-06-23 20:27:36.043058446  [2026-06-23 20:27:36] frigate.app                    INFO    : Review process started: 876
2026-06-23 20:27:36.049124028  [2026-06-23 20:27:36] frigate.app                    INFO    : go2rtc process pid: 129
2026-06-23 20:27:36.558496261  [2026-06-23 20:27:36] frigate.app                    INFO    : Embedding process started: 893
2026-06-23 20:27:36.608526795  [2026-06-23 20:27:36] frigate.detectors.plugins.onnx INFO    : ONNX: loading /config/model_cache/yolov9-s.onnx
2026-06-23 20:27:36.688035906  [2026-06-23 20:27:36] frigate.app                    INFO    : Output process started: 939
2026-06-23 20:27:36.830160922  [2026-06-23 20:27:36] frigate.camera.maintainer      INFO    : Camera processor started for camera_1: 985
2026-06-23 20:27:36.944493432  [2026-06-23 20:27:36] frigate.camera.maintainer      INFO    : Capture process started for camera_1: 1042
2026-06-23 20:27:37.031731288  [2026-06-23 20:27:37] frigate.api.fastapi_app        INFO    : Starting FastAPI app
2026-06-23 20:27:37.285431052  [2026-06-23 20:27:37] frigate.camera.maintainer      INFO    : Camera processor started for camera_2: 1108
2026-06-23 20:27:37.535015395  [2026-06-23 20:27:37] frigate.api.fastapi_app        INFO    : FastAPI started
2026-06-23 20:27:37.589372180  [2026-06-23 20:27:37] frigate.camera.maintainer      INFO    : Capture process started for camera_2: 1209
2026-06-23 20:27:37.674846870  [2026-06-23 20:27:37] frigate.camera.maintainer      INFO    : Camera processor started for camera_3: 1281
2026-06-23 20:27:37.839535295  [2026-06-23 20:27:37] frigate.camera.maintainer      INFO    : Capture process started for camera_3: 1367
2026-06-23 20:27:37.919552637  [2026-06-23 20:27:37] frigate.camera.maintainer      INFO    : Camera processor started for camera_4: 1433
2026-06-23 20:27:38.002233315  [2026-06-23 20:27:38] frigate.camera.maintainer      INFO    : Capture process started for camera_4: 1490
2026-06-23 20:27:38.096863031  [2026-06-23 20:27:38] frigate.camera.maintainer      INFO    : Camera processor started for camera_5: 1557
2026-06-23 20:27:38.117615113  �[0;93m2026-06-23 20:27:38.117453766 [W:onnxruntime:, transformer_memcpy.cc:83 ApplyImpl] 14 Memcpy nodes are added to the graph main_graph for CUDAExecutionProvider. It might have negative impact on performance (including unable to run CUDA graph). Set session_options.log_severity_level=1 to see the detail logs before this message.�[m
2026-06-23 20:27:38.125199048  �[1;31m2026-06-23 20:27:38.125055045 [E:onnxruntime:, inference_session.cc:2094 Initialize] This session cannot use the graph capture feature as requested by the user  as all compute graph nodes have not been partitioned to the CUDAExecutionProvider�[m
2026-06-23 20:27:38.142781216  Process frigate.detector:yolov9_detector:
2026-06-23 20:27:38.142786025  Traceback (most recent call last):
2026-06-23 20:27:38.142787889    File "/usr/lib/python3.11/multiprocessing/process.py", line 314, in _bootstrap
2026-06-23 20:27:38.142788961      self.run()
2026-06-23 20:27:38.142790334    File "/opt/frigate/frigate/object_detection/base.py", line 143, in run
2026-06-23 20:27:38.142791937      object_detector = LocalObjectDetector(detector_config=self.detector_config)
2026-06-23 20:27:38.142798059                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2026-06-23 20:27:38.142811115    File "/opt/frigate/frigate/object_detection/base.py", line 62, in __init__
2026-06-23 20:27:38.142812788      self.detect_api = create_detector(detector_config)
2026-06-23 20:27:38.142814081                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2026-06-23 20:27:38.142831565    File "/opt/frigate/frigate/detectors/__init__.py", line 18, in create_detector
2026-06-23 20:27:38.142832868      return api(detector_config)
2026-06-23 20:27:38.142834080             ^^^^^^^^^^^^^^^^^^^^
2026-06-23 20:27:38.142835583    File "/opt/frigate/frigate/detectors/plugins/onnx.py", line 39, in __init__
2026-06-23 20:27:38.142847487      self.runner = get_optimized_runner(
2026-06-23 20:27:38.142848930                    ^^^^^^^^^^^^^^^^^^^^^
2026-06-23 20:27:38.142850563    File "/opt/frigate/frigate/detectors/detection_runners.py", line 592, in get_optimized_runner
2026-06-23 20:27:38.142851755      ort.InferenceSession(
2026-06-23 20:27:38.142853679    File "/usr/local/lib/python3.11/dist-packages/onnxruntime/capi/onnxruntime_inference_collection.py", line 472, in __init__
2026-06-23 20:27:38.142855373      self._create_inference_session(providers, provider_options, disabled_optimizers)
2026-06-23 20:27:38.142857587    File "/usr/local/lib/python3.11/dist-packages/onnxruntime/capi/onnxruntime_inference_collection.py", line 561, in _create_inference_session
2026-06-23 20:27:38.142871635      sess.initialize_session(providers, provider_options, disabled_optimizers)
2026-06-23 20:27:38.142874661  onnxruntime.capi.onnxruntime_pybind11_state.Fail: [ONNXRuntimeError] : 1 : FAIL : This session cannot use the graph capture feature as requested by the user  as all compute graph nodes have not been partitioned to the CUDAExecutionProvider
2026-06-23 20:27:38.188416120  [2026-06-23 20:27:38] frigate.camera.maintainer      INFO    : Capture process started for camera_5: 1615
2026-06-23 20:27:38.270301487  [2026-06-23 20:27:38] frigate.camera.maintainer      INFO    : Camera processor started for camera_6: 1681
2026-06-23 20:27:38.348034902  [2026-06-23 20:27:38] frigate.camera.maintainer      INFO    : Capture process started for camera_6: 1738
2026-06-23 20:27:38.426915826  [2026-06-23 20:27:38] frigate.camera.maintainer      INFO    : Camera processor started for camera_7: 1806
2026-06-23 20:27:38.572352685  [2026-06-23 20:27:38] frigate.camera.maintainer      INFO    : Capture process started for camera_7: 1863
2026-06-23 20:27:38.646956863  [2026-06-23 20:27:38] frigate.camera.maintainer      INFO    : Camera processor started for camera_8: 1929
2026-06-23 20:27:38.724772099  [2026-06-23 20:27:38] frigate.camera.maintainer      INFO    : Capture process started for camera_8: 1986

Install method

Docker Compose

Object Detector

TensorRT

Screenshots of the Frigate UI's System metrics pages

it wont load properly, only blank boxes and i try to switch the tabs to sorage and network. the whole page goes blank

Any other information that may be helpful

No response

[inkeep-ai-assistant[bot]]

The error indicates that the YOLOv9 model cannot be fully partitioned to the CUDA execution provider, which prevents CUDA graph capture from working. (1)

Your configuration has a few issues:

1. Remove the device: cuda and onnx: block - These are not valid configuration options for the ONNX detector. The GPU is automatically detected when using the -tensorrt image. (1)

2. Simplify your detector configuration - Use this format instead:

detectors:
  onnx:
    type: onnx

model:
  model_type: yolo-generic
  width: 640
  height: 640
  input_tensor: nchw
  input_dtype: float
  path: /config/model_cache/yolov9-s.onnx
  labelmap_path: /labelmap/coco-80.txt

(1)

3. Add the labelmap - Your config is missing labelmap_path: /labelmap/coco-80.txt which is required for YOLO models. (1)

The YOLOv9 model supports CUDA Graphs for optimal Nvidia performance when properly configured. (1) With the -tensorrt Frigate image, Nvidia GPUs will automatically be detected and used with the ONNX detector. (1)

If you continue to have issues, verify that your YOLOv9 model was exported correctly following the export instructions, and ensure your Nvidia driver version is >=545 as required for the 12.x series of CUDA libraries. (2)

📚 Sources:

1. Supported Hardware
2. Recommended hardware

---

Have another question? Just tag @inkeep.

[Pasha-Man]

this is the main issue that i am facing now

[NickM-27]

we usually see this when the model was not exported correctly or a different model was used. What exactly have you run to generate this model?

[Pasha-Man]

i have tried many variations with AI. some say to only use onvx and nvidia will be auto detected. some say to explecitly state it. if the model somehow loads, that the mempy error crashes everything

[NickM-27]

none of that is needed, the only config that is needed is what is in the docs. Like I said above, this is likely to be a model issue, if not then a host driver issue