Unable to install CUDA

Robotics & Edge Computing Jetson & Embedded Systems Jetson Nano
1

Running Jetpack 6.2 L4T 36.4.3

Tried to install PyTorch and the CUDA is is not installing.

python3 -c “import torch; print(‘CUDA:’, torch.cuda.is_available())”
Traceback (most recent call last):
File “”, line 1, in
File “/home/user/.local/lib/python3.10/site-packages/torch/_init_.py”, line 237, in
from torch._C import * # noqa: F403
ImportError: /lib/aarch64-linux-gnu/libcudnn.so.8: version `libcudnn.so.8’ not found (required by /home/coperai/.local/lib/python3.10/site-packages/torch/lib/libtorch_python.so)

Also if I upgrade to 36.4.7

I get this issue! JetPack 36.4.7 GPU Memory Allocation Regression on Orin Nano

Everything seems broken and I will not buy another device if this one can not do what is supposed to do.

2

Hi @alcanjp,

Hope you are doing well!

Just a couple of question to understand the context better:

  1. How did you flash your Jetson board? If you used NVIDIA SDK manager, it automatically installs the correct version of CUDA and cuDNN to match your JetPack release. No need to install them manually.

  2. How did you install PyTorch?

I also found these forums: Torch on Jetson Jetpack 6.2 which may help you install Pytorch.

Best regards,
Nico

4

Hi,

Your PyTorch tries to find cuDNN 8, which means the package is not built for JetPack 6.2.
As JetPack 6.2 includes cuDNN 9.3.0.

Thanks.

5

The issue with CUDA was resolved.
https://ninjalabo.ai/blogs/jetson_pytorch.html

Now this issue:
JetPack 36.4.7 GPU Memory Allocation Regression on Orin Nano. I have to use Jetpack 6.2 L4T 36.4.3

6

Hi,

We are working on CUDA malloc issue when upgrading to the r36.4.7.
Will keep you updated for the status. You can also find more information in the topic below:

Thanks.