[Beta Support]: Nvidia GPU not detected

[skyclaw22]

Describe the problem you are having

The beta2 tensorrt docker image does not auto recognize my gpu to use as detector, and uses CPU
And when manually configuring an onnx detectors frigate crash with:
TypeError: Unable to load from type ‘<class ‘NoneType’>’

Beta Version

0.17.0-beta2

Issue Category

Hardware Acceleration

Frigate config file

default

Relevant Frigate log output

WARNING: cpu detectors are not recommended……….

Relevant go2rtc log output (if applicable)

No response

Install method

Docker Compose

docker-compose file or Docker CLI command

as in docs

Operating system

Ubuntu

CPU / GPU / Hardware

NVIDIA L40S

Screenshots

No response

Steps to reproduce

No response

Any other information that may be helpful

No response

[hawkeye217]

Frigate doesn't auto-detect object detection hardware. It must be specified manually.

As the official documentation indicates, for Nvidia GPUs you need to:

1. Use the -tensorrt Docker image
2. Configure the ONNX detector
3. Download and configure a model for the ONNX detector

If you've already done all of these things, then we'll need to see your full config and the logs in order to help further.

[skyclaw22]

Docs does say in multiple versions that onnx is auto detect for nvidiagpu machine. yet i did miss the part where i must manually download a model for it to work, will try☺️

[hawkeye217]

Can you point out where you see that in the docs for object detection? We want to make the docs as clear as possible.

For enrichments (semantic search, LPR, face recognition) and hardware accelerated video decoding, some GPUs are automatically detected. But this is not the case for object detection.

EDIT: I see where you might be confused. The docs say:

Nvidia GPUs will automatically be detected and used with the ONNX detector in the -tensorrt Frigate image.

The key here is "used with the ONNX detector". You still need to direct your installation to use the ONNX detector - Frigate will not use the ONNX detector by default. Once you configure the ONNX detector and a model, it will correctly use your Nvidia GPU.

[skyclaw22]

that was exactly the place i got confused, just downloaded a model and every thing seems to work, thanks

[hawkeye217]

Great, glad it's working for you now. Happy to help.