CUDA problem, Nvidia jetson AGX Orin

Background: We are students trying to implement vision-control in a vehicle that uses brushless DC motors. We recently received a Nvidia Jetson AGX Orin 32GB H01 kit from SEEED studio to achieve our objective. I want to use YoloP for object detection and lane tracking, but I encountered multiple problems.

Problem: After installing YoloP through GitHub, I tried to run the demo file, but there were issues with CUDA. I tried to check the CUDA version (nvcc --version), but the system responded with command not found. I tried to install Nvidia SDK manager, but the terminal responded with multiple dependencies that are unavailable and not installable.

Downloaded/Installed items: NVIDIA SDK Manager 2.1.0 AMD64.deb (failed to install), OpenCV, YoloP clone repository, Jetpack 5.0.2, Ubuntu 20.04, and PyTorch 2.3.0.

I read that Jetpack includes CUDA, and I installed it through the terminal (sudo apt install nvidia-jetpack) according to Nvidia’s set up guide video.

I am not sure where to go from here, and I would greatly appreciate any help. Thank you for reading!

Fig 1: failure to run YoloP demo code after cloning it from GitHub repository (CUDA issue).

Kind regards

Fig 2: SDK manager failure to install

Hi,

SDK manager is run on a host and flash/setup device through USB.

The error says your PyTorch model doesn’t support CUDA.
How do you install it?
Please try our prebuilt PyTorch package below:

Thanks.

Hi,

apologies for the late reply, I have followed the instructions in the link you had sent me, but there is another error.

Additionally, I have installed SDK manager through Virtualbox Ubuntu 22.04 on a laptop. But after logging in, it says “OOPS no SDKs are available for your account”. I am using a developer account so I would not think there are no SDKs available. I have yet to flash jetpack into the AGX Orin.

Is it likely that the root issue is not flashing the jetson with jetpack?

Thank you.

Hi,

The ‘torch._custom_ops’ issue is more related to the compatibility between PyTorch and TorchVision.

Please note that the version of TorchVision is required to be installed correspondingly.
You can find the compatible version in the below topic:

Thanks.

Hi AastaLLL,

I recently made a new post after I was able to flash the jetson orin. Now I am facing a new issue while trying to use PyTorch.
Could you help me take a look? I would greatly appreciate it.
Link: Pytorch error installing, missing files

Sure. Let’s track the new issue on the new topic and close this one.