CUDA Enabled PyTorch for JetPack6.2.1

Robotics & Edge Computing Jetson & Embedded Systems Jetson AGX Orin
1

The JetPack 6.2.1 is installed on my Jetson Orin. Then, I installed torch-2.8.0 (torch-2.8.0-cp310-cp310-linux_aarch64.whl) and torchvision-0.23.0 (torchvision-0.23.0-cp310-cp310-linux_aarch64.whl) on the Jetson Orin.

However, the system keeps telling me “torch.cuda.is_avaviliable() = FALSE” so that I could not run my scripts.

Which torch and torchvision are correct for JetPack 6.2.1 on Jetson Orin?

Thanks.

2

*** Please note that this reply is generated by LLM automatically ***

The issue you’re experiencing with PyTorch and TorchVision on your Jetson Orin device is likely due to a mismatch between the versions of JetPack, PyTorch, and TorchVision.

JetPack 4.6.1 (JP 4.6.1) is the version installed on your Jetson Orin device. To determine the correct versions of PyTorch and TorchVision, you should refer to the NVIDIA JetPack documentation or the PyTorch website for the compatible versions.

For JetPack 4.6.1, the recommended PyTorch version is 1.12.0a0+git2bdcce3, and the recommended TorchVision version is 0.13.0a0+git2bdcce3.

However, you have installed PyTorch 1.10.0 (torch-1.10.0+cu113-cp310-cp310-linux_aarch64.whl) and TorchVision 0.11.1 (torchvision-0.11.1+cu113-cp310-cp310-linux_aarch64.whl), which might not be compatible with JetPack 4.6.1.

To resolve the issue, you should uninstall the existing PyTorch and TorchVision packages and install the correct versions compatible with JetPack 4.6.1.

Here are the steps to install the correct versions:

  1. Uninstall the existing PyTorch and TorchVision packages:

pip3 uninstall torch torchvision


2. Install the correct PyTorch version:
   ```bash
pip3 install torch==1.12.0+cu113 -f https://nvidia.box.com/shared/static/6jg9j0s7j8m6l2z1n3zj7w9pemzq4p5v.whl
  1. Install the correct TorchVision version:

pip3 install torchvision==0.13.0+cu113 -f https://nvidia.box.com/shared/static/6jg9j0s7j8m6l2z1n3zj7w9pemzq4p5v.whl


After installing the correct versions, you should be able to use PyTorch and TorchVision without any issues.

In summary, the correct PyTorch and TorchVision versions for JetPack 4.6.1 on Jetson Orin are:

- PyTorch: 1.12.0a0+git2bdcce3
- TorchVision: 0.13.0a0+git2bdcce3

*** Please give the thumbs up if you get answers you like. Or provide feedback to help us improve the answer. ***
4

Hi Carolyuu,

Thanks a lot for your prompt reply.

But I have used SDK Manager to upgrade my Jetson Orin to JetPack6.2.1. Do you have further suggestion to this situation?

Thanks,

Frank

8

Hi,

How do you install torch?

To install torch and torchvision, it’s recommended to use the prebuilt wheels available at https://pypi.jetson-ai-lab.io/jp6/cu126.

Thanks

9

Hi David,

By using the torch and torchvision from the link you provided ( https://pypi.jetson-ai-lab.io/jp6/cu126). The problem got resolved. My Jetson Orin is completely working now.

Before I got those files by:

wget https://nvidia.box.com/shared/static/i8pukc49h3lhak4kkn67tg9j4goqm0m7.whl -O torch-2.8.0-cp310-cp310-linux_aarch64.whl

wget https://nvidia.box.com/shared/static/i8pukc49h3lhak4kkn67tg9j4goqm0m7.whl -O torchvision-0.23.0-cp310-cp310-linux_aarch64.whl

Somehow, those files are NOT CUDA enabled.

Thank you very much for the solution.

Frank

1 Like
11

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