[HW Accel Support]: Docker Compose and Installation with NVIDIA 5070 #18169

angeres1 · 2025-05-11T19:26:26Z

angeres1
May 11, 2025

Describe the problem you are having

When trying to deploy Frigate inside a VM in Proxmox I am getting:
Using: ghcr.io/blakeblackshear/frigate:stable-tensorrt

Version

frigate:stable-tensorrt

Frigate config file

# Frigate Configuration File (config.yml)

# Base configuration - MQTT, Database, Detectors

mqtt:

  enabled: True # Set to True for Home Assistant integration

  host: 192.168.10.* # IP of your Home Assistant VM (which should run an MQTT broker)

  port: 1883

  user: "user_mqtt_user" # Replace with your MQTT username

  password: "pass_mqtt_password" # Replace with your MQTT password

  # topic_prefix: frigate # Default, can be changed if needed



database:

  path: /media/frigate/frigate.db # Path inside the container, maps to NAS



# Available detectors (NVIDIA will be primary)

detectors:

  cpu_detector: # A fallback CPU detector (optional, but good to have)

    type: cpu

  # Coral EdgeTPU detector (if you ever add one)

  # coral_detector:

  #   type: edgetpu

  #   device: usb # or pci depending on your Coral

  # NVIDIA GPU Detector

  # For RTX series, TensorRT is generally preferred for best performance.

  # Check Frigate documentation for the most up-to-date NVIDIA detector config.

  # It usually relies on the NVIDIA Container Toolkit and drivers being present.

  main_nvidia_detector:

    type: nvidia

    # device: 0 # Usually the first GPU passed to the container. Frigate often auto-de>

    # For newer Frigate versions, specific tensorrt options might be available or set >



# FFmpeg settings (usually fine with defaults unless troubleshooting)

# ffmpeg:

  # Optional: Global FFmpeg hardware acceleration arguments (can be overridden per cam>

  # hwaccel_args: preset-nvidia-h264 # Or preset-nvidia-hevc if your cameras use H.265>

  # preset-nvidia-h264 typically includes:

  # - -hwaccel

  # - nvidia

  # - -hwaccel_output_format

  # - cuda

  # - -c:v

  # - h264_cuvid # or hevc_cuvid

  # Ensure your ffmpeg build inside Frigate container supports this.

  # Frigate's official images usually have good NVIDIA support.

#  pass_through_args: [] # Default



# Model (Object detection model - Frigate usually defaults to a good one)

model:

  width: 320 # Default input size for detection model

  height: 320 # Default input size for detection model

  # path: /cpu_models/ssdlite_mobilenet_v2.tflite # Example, Frigate manages this



# Object tracking and filters

objects:

  track: # List of objects to track

    - person

    - car

    - cat

    - dog

    # Add other objects as needed (e.g., bicycle, truck, etc.)

  filters: # Optional: filters per object type

    person:

      min_area: 5000 # Minimum pixel area to consider a detection valid

      max_area: 100000

      threshold: 0.6 # Minimum confidence score (0.0-1.0)

    car:

      min_area: 10000

      threshold: 0.5



# Record settings (global, can be overridden per camera)

record:

  enabled: True

  retain:

    days: 7

    mode: motion

  events: # Events config lives under record

    retain: # Retain config for events lives under events

      default: 10 # How long to keep event data

      # mode: motion # Mode for events is determined by detect logic, not set here exp>



# Snapshots settings (global, can be overridden per camera)

snapshots:

  enabled: True

  timestamp: False # Add timestamp to snapshots

  bounding_box: True # Draw bounding box on snapshots

  retain:

    default: 14 # Days to keep snapshots



# Optional: Birdseye view

# birdseye:

#   enabled: True # False, Restream, Transcode

#   width: 1280

#   height: 720

#   quality: 8

#   mode: objects # 'objects' or 'motion'



# Go2RTC configuration (for WebRTC, MSE, etc. - Frigate manages this by default)

# Usually no need to configure this unless you have advanced needs.



# Camera specific configurations

cameras:

  # Replace with your actual camera configurations

  # Example for one Reolink RLC-810A camera:

  # Give each camera a unique name (no spaces, underscores allowed)

  living_room_cam: # << UNIQUE CAMERA NAME

    ffmpeg:

      inputs:

        # Input for detection (usually a lower-resolution substream)

        - path: rtsp://user:[email protected].*:554/h264Previ>

          roles:

            - detect

        # Input for recording/streaming (usually the high-resolution main stream)

        - path: rtsp://user:[email protected].*:554/h264Previ>

          roles:

            - record

            # - rtmp # If you want to restream via RTMP

    detect: # Detection settings for this camera

      enabled: True

      width: 704 # Substream width (adjust to your Reolink RLC-810A substream resoluti>

      height: 480 # Substream height (adjust to your Reolink RLC-810A substream resolu>

      fps: 5 # FPS for detection processing

      # detector: main_nvidia_detector # Explicitly assign detector if needed, otherwi>

    record: # Recording settings for this camera (overrides global)

      enabled: True

      # retain: # Override global retain if needed

      #   days: 10

    snapshots: # Snapshot settings for this camera (overrides global)

      enabled: True

      # retain: # Override global retain if needed

      #   default: 30

    # Optional: Define zones for this camera

    # zones:

    #   driveway_zone:

    #     coordinates: 0,0,100,0,100,100,0,100 # Example: top-left, top-right, bottom->

    #     objects:

    #       - person

    #       - car

    # Optional: Motion masks for this camera

    # motion:

    #   mask:

    #     - 0,0,100,0,100,10,0,10 # Example mask for the top 10 pixels



  # Add more cameras here, copying the structure above:

  # another_cam:

  #   ffmpeg:

  #     inputs:

  #       - path: rtsp://...

  #         roles:

  #           - detect

  #       - path: rtsp://...

  #         roles:

  #           - record

  #   detect:

  #     ...

  #   ...

docker-compose file or Docker CLI command

version: '3.9' # This line can be removed
services:
  frigate:
    container_name: frigate
    privileged: true
    restart: unless-stopped
    image: ghcr.io/blakeblackshear/frigate:stable-tensorrt
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1 # Or 'all'
              capabilities: [gpu] # Correct placement for device capabilities
    volumes:
      - /etc/localtime:/etc/localtime:ro
      - ./config.yml:/config/config.yml:ro
      - /mnt/psmfrigate_recordings:/media/frigate
      - type: tmpfs
        target: /tmp/cache
        tmpfs:
          size: 1000000000
    ports:
      - "5000:5000"
      - "8971:8971"
      - "8554:8554"
      - "8555:8555/tcp"
      - "8555:8555/udp"
    environment:
      - NVIDIA_DRIVER_CAPABILITIES=all
      - NVIDIA_VISIBLE_DEVICES=all
     # - LD_LIBRARY_PATH=/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/lib/x86_64-linux-gnu:/usr/local/lib/python3.9/dist-packages/tensorrt # Try with this commented out first

Relevant Frigate log output

WARN[0000] /opt/stacks/frigate/docker-compose.yml: the attribute `version` is obsolete, it will be ignored, please remove it to avoid potential confusion
frigate  | s6-rc: info: service s6rc-fdholder: starting
frigate  | s6-rc: info: service s6rc-oneshot-runner: starting
frigate  | s6-rc: info: service s6rc-oneshot-runner successfully started
frigate  | s6-rc: info: service fix-attrs: starting
frigate  | s6-rc: info: service s6rc-fdholder successfully started
frigate  | s6-rc: info: service fix-attrs successfully started
frigate  | s6-rc: info: service legacy-cont-init: starting
frigate  | s6-rc: info: service legacy-cont-init successfully started
frigate  | s6-rc: info: service trt-model-prepare: starting
frigate  | s6-rc: info: service log-prepare: starting
frigate  | tensorrt model preparation disabled
frigate  | s6-rc: info: service trt-model-prepare successfully started
frigate  | s6-rc: info: service log-prepare successfully started
frigate  | s6-rc: info: service nginx-log: starting
frigate  | s6-rc: info: service go2rtc-log: starting
frigate  | s6-rc: info: service frigate-log: starting
frigate  | s6-rc: info: service certsync-log: starting
frigate  | s6-rc: info: service nginx-log successfully started
frigate  | s6-rc: info: service frigate-log successfully started
frigate  | s6-rc: info: service go2rtc-log successfully started
frigate  | s6-rc: info: service go2rtc: starting
frigate  | s6-rc: info: service certsync-log successfully started
frigate  | s6-rc: info: service go2rtc successfully started
frigate  | s6-rc: info: service go2rtc-healthcheck: starting
frigate  | s6-rc: info: service frigate: starting
frigate  | s6-rc: info: service go2rtc-healthcheck successfully started
frigate  | s6-rc: info: service frigate successfully started
frigate  | s6-rc: info: service nginx: starting
frigate  | 2025-05-11 19:15:44.070005401  [INFO] Preparing new go2rtc config...
frigate  | 2025-05-11 19:15:44.070320317  [INFO] Preparing Frigate...
frigate  | 2025-05-11 19:15:44.071051281  [INFO] Starting NGINX...
frigate  | 2025-05-11 19:15:44.090293442  [INFO] No TLS certificate found. Generating a self signed certificate...
frigate  | 2025-05-11 19:15:44.187736153  [INFO] Starting Frigate...
frigate  | 2025-05-11 19:15:44.444089621  [INFO] Starting go2rtc...
frigate  | 2025-05-11 19:15:44.498677234  19:15:44.498 INF go2rtc platform=linux/amd64 revision=b2399f3 version=1.9.2
frigate  | 2025-05-11 19:15:44.498681141  19:15:44.498 INF config path=/dev/shm/go2rtc.yaml
frigate  | 2025-05-11 19:15:44.498947004  19:15:44.498 INF [rtsp] listen addr=:8554
frigate  | 2025-05-11 19:15:44.499230692  19:15:44.499 INF [webrtc] listen addr=:8555/tcp
frigate  | 2025-05-11 19:15:44.499232014  19:15:44.499 INF [api] listen addr=:1984
frigate  | s6-rc: info: service nginx successfully started
frigate  | s6-rc: info: service certsync: starting
frigate  | s6-rc: info: service certsync successfully started
frigate  | s6-rc: info: service legacy-services: starting
frigate  | 2025-05-11 19:15:45.582687839  [INFO] Starting certsync...
frigate  | s6-rc: info: service legacy-services successfully started
frigate  | 2025-05-11 19:15:45.619556755  127.0.0.1 - - [11/May/2025:19:15:45 +0000] "" 400 0 "-" "-" "-"
frigate  | 2025-05-11 19:15:46.259918125  [2025-05-11 19:15:46] frigate.util.config            INFO    : Checking if frigate config needs migration...
frigate  | 2025-05-11 19:15:46.259932993  [2025-05-11 19:15:46] frigate.util.config            ERROR   : Config file is read-only, unable to migrate config file.
frigate  | 2025-05-11 19:15:47.381644415  [2025-05-11 19:15:47] frigate.util.services          INFO    : Automatically detected nvidia hwaccel for video decoding
frigate  | 2025-05-11 19:15:47.384257181  [2025-05-11 19:15:47] frigate.app                    INFO    : Starting Frigate (0.15.0-6cb5cfb)
frigate  | 2025-05-11 19:15:47.405814117  [2025-05-11 19:15:47] peewee_migrate.logs            INFO    : Starting migrations
frigate  | 2025-05-11 19:15:47.407477967  [2025-05-11 19:15:47] peewee_migrate.logs            INFO    : There is nothing to migrate
frigate  | 2025-05-11 19:15:47.410102907  [2025-05-11 19:15:47] frigate.app                    INFO    : Running database vacuum
frigate  | 2025-05-11 19:15:47.433150284  [2025-05-11 19:15:47] frigate.app                    INFO    : Recording process started: 406
frigate  | 2025-05-11 19:15:47.433426236  [2025-05-11 19:15:47] frigate.app                    INFO    : Review process started: 413
frigate  | 2025-05-11 19:15:47.434683707  [2025-05-11 19:15:47] frigate.app                    INFO    : go2rtc process pid: 103
frigate  | 2025-05-11 19:15:47.451782451  [2025-05-11 19:15:47] frigate.app                    INFO    : Output process started: 438
frigate  | 2025-05-11 19:15:47.471120355  [2025-05-11 19:15:47] frigate.app                    INFO    : Camera processor started for test_camera_1: 459
frigate  | 2025-05-11 19:15:47.471123130  [2025-05-11 19:15:47] frigate.app                    INFO    : Capture process started for test_camera_1: 461
frigate  | 2025-05-11 19:15:47.489869243  [2025-05-11 19:15:47] detector.main_nvidia_detector  INFO    : Starting detection process: 422
frigate  | 2025-05-11 19:15:47.565785650  Process detector:main_nvidia_detector:
frigate  | 2025-05-11 19:15:47.565788094  Traceback (most recent call last):
frigate  | 2025-05-11 19:15:47.565788846    File "/usr/lib/python3.9/multiprocessing/process.py", line 315, in _bootstrap
frigate  | 2025-05-11 19:15:47.565789267      self.run()
frigate  | 2025-05-11 19:15:47.565789737    File "/opt/frigate/frigate/util/process.py", line 41, in run_wrapper
frigate  | 2025-05-11 19:15:47.565791050      return run(*args, **kwargs)
frigate  | 2025-05-11 19:15:47.565793355    File "/usr/lib/python3.9/multiprocessing/process.py", line 108, in run
frigate  | 2025-05-11 19:15:47.565793796      self._target(*self._args, **self._kwargs)
frigate  | 2025-05-11 19:15:47.565794236    File "/opt/frigate/frigate/object_detection.py", line 121, in run_detector
frigate  | 2025-05-11 19:15:47.565804906      object_detector = LocalObjectDetector(detector_config=detector_config)
frigate  | 2025-05-11 19:15:47.565805418    File "/opt/frigate/frigate/object_detection.py", line 68, in __init__
frigate  | 2025-05-11 19:15:47.565805848      self.detect_api = create_detector(detector_config)
frigate  | 2025-05-11 19:15:47.565818893    File "/opt/frigate/frigate/detectors/__init__.py", line 18, in create_detector
frigate  | 2025-05-11 19:15:47.565819313      return api(detector_config)
frigate  | 2025-05-11 19:15:47.565829171    File "/opt/frigate/frigate/detectors/plugins/tensorrt.py", line 243, in __init__
frigate  | 2025-05-11 19:15:47.565829612      self.engine = self._load_engine(detector_config.model.path)
frigate  | 2025-05-11 19:15:47.565830124    File "/opt/frigate/frigate/detectors/plugins/tensorrt.py", line 87, in _load_engine
frigate  | 2025-05-11 19:15:47.565830574      with open(model_path, "rb") as f, trt.Runtime(self.trt_logger) as runtime:
frigate  | 2025-05-11 19:15:47.565831005  TypeError: expected str, bytes or os.PathLike object, not NoneType
frigate  | 2025-05-11 19:15:47.565831426  Exception ignored in: <function TensorRtDetector.__del__ at 0x79f05df15c10>
frigate  | 2025-05-11 19:15:47.565831776  Traceback (most recent call last):
frigate  | 2025-05-11 19:15:47.565842797    File "/opt/frigate/frigate/detectors/plugins/tensorrt.py", line 262, in __del__
frigate  | 2025-05-11 19:15:47.565843209      if self.outputs is not None:
frigate  | 2025-05-11 19:15:47.565843619  AttributeError: 'TensorRtDetector' object has no attribute 'outputs'
frigate  | 2025-05-11 19:15:47.778580096  [2025-05-11 19:15:47] frigate.api.fastapi_app        INFO    : Starting FastAPI app
frigate  | 2025-05-11 19:15:47.822120264  [2025-05-11 19:15:47] frigate.api.fastapi_app        INFO    : FastAPI started

Relevant go2rtc log output

N/A

FFprobe output from your camera

N/A

Install method

Proxmox via Docker

Object Detector

TensorRT

Network connection

Wired

Camera make and model

Reolink RLC-810A

Screenshots of the Frigate UI's System metrics pages

N/A

Any other information that may be helpful

Answered by hawkeye217

May 12, 2025

Frigate 0.16 uses a slightly newer trt version, but in general if that doesn't work then you are talking about a major version upgrade to tensorrt 10 which is going to be a lot of refactoring.

Realistically though, what you should be able to do instead is just use 0.16 and then you can generate the onnx model (instead of converting to tensorrt) and run via onnx which should work fine on the 5070. If you do decide to go this way keep in mind that 0.16 is still in beta. There are dev specific docs which walk through how to download these models:

https://deploy-preview-16390--frigate-docs.netlify.app/configuration/object_detectors#onnx

View full answer

hawkeye217 · 2025-05-11T19:28:37Z

hawkeye217
May 11, 2025
Collaborator

You need to rebuild your trt-models folder. See the documentation: https://docs.frigate.video/configuration/object_detectors/#nvidia-tensorrt-detector

8 replies

angeres1 May 11, 2025
Author

Tried with:
• YOLO_MODELS=yolov7-320,yolov7x-640
• USE_FP16=true
• TRT_MODEL_PREP_DEVICE=0

But:
Creating yolov7-320.cfg and yolov7-320.weights
Creating yolov7x-640.cfg and yolov7x-640.weights

Done.

Generating yolov7-320.trt. This may take a few minutes.

Traceback (most recent call last):
File "/usr/local/src/tensorrt_demos/yolo/onnx_to_tensorrt.py", line 244, in
main()
File "/usr/local/src/tensorrt_demos/yolo/onnx_to_tensorrt.py", line 229, in main
engine = build_engine(
File "/usr/local/src/tensorrt_demos/yolo/onnx_to_tensorrt.py", line 114, in build_engine
with trt.Builder(TRT_LOGGER) as builder, builder.create_network(*EXPLICIT_BATCH) as network, trt.OnnxParser(network, TRT_LOGGER) as parser:
TypeError: pybind11::init(): factory function returned nullptr
[05/11/2025-23:50:58] [TRT] [E] 1: [caskUtils.cpp::createCaskHardwareInfo::857] Error Code 1: Cask (no further information)
[05/11/2025-23:50:58] [TRT] [E] 1: [checkMacros.cpp::catchCudaError::202] Error Code 1: Cuda Runtime (no kernel image is available for execution on the device)
Loading the ONNX file...

Generating yolov7x-640.trt. This may take a few minutes.

Traceback (most recent call last):
File "/usr/local/src/tensorrt_demos/yolo/onnx_to_tensorrt.py", line 244, in
main()
File "/usr/local/src/tensorrt_demos/yolo/onnx_to_tensorrt.py", line 229, in main
engine = build_engine(
File "/usr/local/src/tensorrt_demos/yolo/onnx_to_tensorrt.py", line 114, in build_engine
with trt.Builder(TRT_LOGGER) as builder, builder.create_network(EXPLICIT_BATCH) as network, trt.OnnxParser(network, TRT_LOGGER) as parser:
TypeError: pybind11::init(): factory function returned nullptr
[05/11/2025-23:51:05] [TRT] [E] 1: [caskUtils.cpp::createCaskHardwareInfo::857] Error Code 1: Cask (no further information)
[05/11/2025-23:51:05] [TRT] [E] 1: [checkMacros.cpp::catchCudaError::202] Error Code 1: Cuda Runtime (no kernel image is available for execution on the device)
Loading the ONNX file...
Available tensorrt models:
ls: cannot access '.trt': No such file or directory
s6-rc: warning: unable to start service trt-model-prepare: command exited 2

ChatGPT is telling me:
Frigate tries to generate the .trt models internally using scripts that are compiled for older GPUs (Ampere, Turing, etc.). The tool onnx_to_tensorrt.py fails because:
• The CUDA/TensorRT runtime does not support the Compute Capability 8.9/9.0+ of the RTX 5070.
• This is a limitation of the current Frigate container (stable-tensorrt), which is based on a pre-packaged version of TensorRT.

The Frigate community has not yet adapted its scripts to TensorRT 9.x or Compute Capability 8.9+ (Ada GPUs). You can follow the official repo to know when it gets updated.

hawkeye217 May 12, 2025
Collaborator

Frigate 0.16 uses a slightly newer trt version, but in general if that doesn't work then you are talking about a major version upgrade to tensorrt 10 which is going to be a lot of refactoring.

Realistically though, what you should be able to do instead is just use 0.16 and then you can generate the onnx model (instead of converting to tensorrt) and run via onnx which should work fine on the 5070. If you do decide to go this way keep in mind that 0.16 is still in beta. There are dev specific docs which walk through how to download these models:

https://deploy-preview-16390--frigate-docs.netlify.app/configuration/object_detectors#onnx

Answer selected by NickM-27

angeres1 May 12, 2025
Author

Hi hawkeye217,
Thanks for your guidance, I made it work!!!
I had to create a new yolov9-t-480x640.onnx with the parameters "inverted" usign:
python export.py --weights ./yolov9-t.pt --imgsz 480 640 --simplify --include onnx
python3 -m onnxsim yolov9-t.onnx yolov9-t-480x640.onnx
Then setting up the config.yml file as:
...
model:
model_type: yolo-generic
width: 640 # Match ONNX shape (W=640)
height: 480 # Match ONNX shape (H=480)
input_tensor: nchw
input_dtype: float
path: /models/yolov9-t-480x640.onnx
labelmap_path: /labelmap/coco-80.txt

cameras:
living_room_cam:
...
detect:
enabled: true
width: 640
height: 480
fps: 5
...

And now it is working, Frigate is detecting the GPU

{"0":{"name":"NVIDIA GeForce RTX 5070","driver":"570.133.07","cuda_compute":12,"vbios":"98.05.28.00.2A"}}

And it is currently using it.

Final question: everytime I am opening the service in my URL: http://192.168.10.*:5000/, it is taking me directly to the main page, it is like bypassing the login page, is this normal? o should I set something else?

NickM-27 May 12, 2025
Collaborator Sponsor

Final question: everytime I am opening the service in my URL: http://192.168.10.*:5000/, it is taking me directly to the main page, it is like bypassing the login page, is this normal? o should I set something else?

yes, port 5000 is unauthenticated

hawkeye217 May 12, 2025
Collaborator

Great! Glad it's working for you.

angeres1 May 12, 2025
Author

Is there a port "authenticated"?

banthungprong · 2025-05-12T07:14:14Z

banthungprong
May 12, 2025

Sorry, but I had to smile:
Proxmox and forwarding problems with ChatGPT to the non-adapting Frigate community.
This is the future.

1 reply

angeres1 May 12, 2025
Author

Yeah!!!
It reached to a point where It was always telling me to wait until a further release... I had to restarted it...
But if works for anyone, GPT was helpfull, I mean, I had to read through the articles and try to guide it through the context I saw It could work and finally we made it.

Uh oh!

[HW Accel Support]: Docker Compose and Installation with NVIDIA 5070 #18169

Uh oh!

angeres1 May 11, 2025

Describe the problem you are having

Version

Frigate config file

docker-compose file or Docker CLI command

Relevant Frigate log output

Relevant go2rtc log output

FFprobe output from your camera

Install method

Object Detector

Network connection

Camera make and model

Screenshots of the Frigate UI's System metrics pages

Any other information that may be helpful

Replies: 2 comments · 9 replies

Uh oh!

hawkeye217 May 11, 2025 Collaborator

Uh oh!

Uh oh!

angeres1 May 11, 2025 Author

Uh oh!

hawkeye217 May 12, 2025 Collaborator

Uh oh!

angeres1 May 12, 2025 Author

Uh oh!

NickM-27 May 12, 2025 Collaborator Sponsor

Uh oh!

hawkeye217 May 12, 2025 Collaborator

Uh oh!

angeres1 May 12, 2025 Author

Uh oh!

banthungprong May 12, 2025

Uh oh!

angeres1 May 12, 2025 Author

angeres1
May 11, 2025

Replies: 2 comments 9 replies

hawkeye217
May 11, 2025
Collaborator

angeres1 May 11, 2025
Author

hawkeye217 May 12, 2025
Collaborator

angeres1 May 12, 2025
Author

NickM-27 May 12, 2025
Collaborator Sponsor

hawkeye217 May 12, 2025
Collaborator

angeres1 May 12, 2025
Author

banthungprong
May 12, 2025

angeres1 May 12, 2025
Author