Troubleshooting Torch CUDA Availability Issue with ZED Box Orin NX

I recently purchased a ZED Box Orin NX and have been conducting tests.

However, I encountered an issue when importing torch in Python. Specifically, the result of ‘torch.cuda.is_available()’ returns false. It seems odd because other functionalities, such as SDK diagnostics, depth viewer, and object tracking, are operating in real-time, suggesting that CUDA is being utilized effectively. I have also confirmed that the CUDA folder is installed.

Despite this, I can’t understand why CUDA isn’t working with torch in Python. I’ve tried setting the environment variables as well. Could you provide insights or solutions to this issue?"

Hi @kyuwha,

First of all, welcome to the Stereolabs community!

Could you please share your ZED_Diagnostic file, as well as your python version and python packages versions? (If you use pip / conda, can you please share the output of pip freeze / conda list)

There may be a mismatch between the cuda version your pytorch version is expecting. Can you validate that your pytorch version is compatible with CUDA installed on your box?