Nvidia Jetson Xavier NX problem installing CUDA and activating GPU

[ExailDeveloper]

I’m working with an Nvidia Jetson Xavier NX for object recognition and, to optimize the process, I’d like to use the GPU, but I’m having problems. I flashed and configured the card using a LInux 20.04 Host PC with the SDK Manager and installed version 5.1 of JetPack along with DeepStream 6.2. (In version 5.1.3 of Jetpack, after sudo apt upgrade, it was no longer possible to boot).I followed the same steps as in this tutorial for configuration: https://www.youtube.com/watch?v=Ucg5Zqm9ZMk

Afterwards, I installed the CUDA 12.1 toolkit, as indicated below, and then pytorch.

• CUDA: Cuda 12.1 driver for Linux aarch64 missing? - #3 by AastaLLL
• Pytorch: Start Locally | PyTorch

However, even though I updated the export path, when I run the nvcc --version command, it shows that the installed version of CUDA is 11.4. Also, when I call the torch.cuda.is_available() function, I get False.

I’d like to know what you need to do to actually use the GPU during testing.

[AastaLLL]

Hi,

For nvcc, please try the below command:

export PATH=/usr/local/cuda/12.1/bin:$PATH
export LD_LIBRARY_PATH=/usr/local/cuda/12.1//lib64:$LD_LIBRARY_PATH

Since you have a custom CUDA and Python version, you will need to build the PyTorch from the source to get the GPU support.

Thanks.

[ExailDeveloper]

Thank you very much, that solved the nvcc --version problem. However, when I run the jtop command, the version of opencv installed is still wrong, it shows 4.5.4 instead of 4.10. Can you tell me what I can do in this case?