Jetson Orin Nx, Pytorch 2.1.2 on Jetpack 6.0 DP, torch.cuda.is_available() is false

Autonomous Machines Jetson & Embedded Systems Jetson Orin NX
1

Hi all,
My hardware is Jetson Orin NX, and I have installed JetPack 6.0 DP. I created a virtual environment using Python 3.8.17 and used the command conda install pytorch torchvision torchaudio pytorch-cuda=12.1 -c pytorch -c nvidia to install PyTorch version 2.1.2.

In the Python interpreter, when using torch.__version__ , it returns ‘2.1.2’. However, when using torch.cuda.is_available() , it returns false. How can I make torch.cuda.is_available() return true?

image
image728×323 6.76 KB

Thank you

3

Hi,

The packages on the apt/pip server are only built with CPU support.
Please use our prebuilt instead to get the CUDA support:

Thanks.

1 Like
4

Hi AastaLLL,

I use Jetpack 6.0 DP，and Pytorch 2.2.0

export TORCH_INSTALL=https://developer.download.nvidia.cn/compute/redist/jp/v60dp/pytorch/torch-2.2.0a0+6a974be.nv23.11-cp310-cp310-linux_aarch64.whl

it work

Thank you

6

Hi AastaLLL,

I install Pytorch 2.2.0,

and I need to install which version of torchvision.

image

7

Hi,

Please try TorchVision 0.17.

Thanks.

8

Hi AasTaLLL,

https://download.pytorch.org/whl/nightly/torchvision/

I download torchvision 0.17 from this website

and use command $ sudo pip3 install torchvision-0.17.0.dev20231211-cp310-cp310-linux_aarch64.whl

but return error message :

ERROR: Could not find a version that satisfies the requirement torch==2.3.0.dev20231211 (from torchvision) (from versions: 1.10.2, 1.11.0, 1.12.0, 1.12.1, 1.13.0, 1.13.1, 2.0.0, 2.0.1, 2.1.0, 2.1.1, 2.1.2)
ERROR: No matching distribution found for torch==2.3.0.dev20231211

and i try torchvision-0.17.0.dev20231010-cp310-cp310-linux_aarch64.whl

return ERROR: No matching distribution found for torch==2.2.0.dev20231010

How can I fix it?

Thanks

9

@cho56915 these wheels from PyTorch.org website were not built with CUDA enabled, they are for general ARM64 platform without GPU. Instead, please build torchvision from source following the torchvision instructions under the Installation section of this topic:

Or you can use this container which already has it:

That has PyTorch/torchvision/torchaudio/ect already installed, and the compiled wheels are saved under /opt inside the container if you want to copy them out.

10

Hi dusty_nv

The issue has been resolved.

Thank you for your suggestions.

12

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.