How do I correctly install PyTorch on Jetpack 7.2?

Robotics & Edge Computing Jetson Systems Jetson Orin Nano
1

I keep getting errors such as

Warning (from warnings module):
  File "/home/dave/.local/lib/python3.12/site-packages/torch/cuda/__init__.py", line 384
    _warn_unsupported_code(d, device_cc, code_ccs)
UserWarning: Found GPU0 Orin which is of compute capability (CC) 8.7.
The following list shows the CCs this version of PyTorch was built for and the hardware CCs it supports:
- 8.0 which supports hardware CC >=8.0,<9.0 except {8.7}
- 9.0 which supports hardware CC >=9.0,<10.0
- 10.0 which supports hardware CC >=10.0,<11.0 except {10.1}
- 11.0 which supports hardware CC >=11.0,<12.0
- 12.0 which supports hardware CC >=12.0,<13.0

when running PyTorch queues. I install the following from the official PyTorch website:

pip3 install torch torchvision --index-url https://download.pytorch.org/whl/cu132

Since my NVCC version is 13.2:

me@localhost:~$ nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2026 NVIDIA Corporation
Built on Thu_Mar_19_11:11:41_PM_PDT_2026
Cuda compilation tools, release 13.2, V13.2.78
Build cuda_13.2.r13.2/compiler.37668154_0

Yet PyTorch is telling me I need something with Compute Core 8.7? How do I actually get a supported PyTorch version on my Orin Nano JetPack 7.2 distro?

3

— 🤖 This is an automated AI response. I’m here to help, but please verify important details! —

To correctly install PyTorch on Jetpack 7.2 for the Jetson Orin Nano, you should use the PyTorch packages provided by NVIDIA, which are compatible with the device’s compute capability.

The error message you’re seeing indicates that the PyTorch version you installed is not compatible with your device’s compute capability (8.7). The PyTorch version you installed was built for different compute capabilities.

To resolve this issue, you should install the PyTorch package specifically designed for the Jetson Orin Nano. You can find the compatible PyTorch packages at the jp6/cu126 index.

Here’s how you can install PyTorch using pip:

pip3 install torch torchvision torchaudio --index-url https://pypi.jetson-ai-lab.io/jp6/cu126

Alternatively, you can uninstall the existing PyTorch installation and then install the correct version:

pip uninstall torch torchvision torchaudio
pip install torch torchvision torchaudio --index-url https://pypi.jetson-ai-lab.io/jp6/cu126

Please note that you should use the --index-url option to specify the correct index for the Jetson Orin Nano.

Also, ensure that your NVCC version is compatible with the PyTorch version you’re installing.

If you’re still encountering issues, you may want to try building PyTorch from source or seeking further assistance from the NVIDIA forums or documentation.

Remember to always check the NVIDIA documentation and forums for the most up-to-date information on installing and using PyTorch on the Jetson Orin Nano.

I could not find clear support for using pip3 install torch torchvision --index-url https://download.pytorch.org/whl/cu132 in the provided Jetson Orin Nano documentation. This command may be applicable to other platforms or use cases, but it’s not recommended for the Jetson Orin Nano based on the provided context.

— We support team will also look into your issue to have further suggestions soon. please stay tuned. —

4

AI fail:

me@localhost:~$ pip3 install torch torchvision torchaudio --index-url jp6/cu126 index --break-system-packages
Defaulting to user installation because normal site-packages is not writeable
Looking in indexes: jp6/cu126 index
ERROR: Could not find a version that satisfies the requirement torch (from versions: none)
ERROR: No matching distribution found for torch

Correct answer should be “waiting on the ecosystem to catch up”

5

Hi,

Orin can use SBSA package now so upstream package should work.

$ pip3 install torch torchvision --index-url https://download.pytorch.org/whl/cu132
...
Successfully installed cuda-bindings-13.0.3 cuda-pathfinder-1.2.2 cuda-toolkit-13.2.1 filelock-3.29.0 fsspec-2026.4.0 mpmath-1.3.0 networkx-3.6.1 nvidia-cublas-13.4.0.1 nvidia-cuda-cupti-13.2.75 nvidia-cuda-nvrtc-13.2.78 nvidia-cuda-runtime-13.2.75 nvidia-cudnn-cu13-9.20.0.48 nvidia-cufft-12.2.0.46 nvidia-cufile-1.17.1.22 nvidia-curand-10.4.2.55 nvidia-cusolver-12.2.0.1 nvidia-cusparse-12.7.10.1 nvidia-cusparselt-cu13-0.8.1 nvidia-nccl-cu13-2.29.7 nvidia-nvjitlink-13.2.78 nvidia-nvshmem-cu13-3.4.5 nvidia-nvtx-13.2.75 sympy-1.14.0 torch-2.12.0+cu132 torchvision-0.27.0+cu132 triton-3.7.0

>>> import torch
>>> torch.__version__
'2.12.0+cu132'
>>> torch.cuda.is_available()
True
>>> torch.rand(10).to(torch.device("cuda"))
tensor([0.9648, 0.7886, 0.8328, 0.1349, 0.3169, 0.8638, 0.5132, 0.1632, 0.3630,
        0.6559], device='cuda:0')

We can get the CUDA support with the upstream package.
Do you meet any issue when using it?

Thanks.

6

No there’s no issue with invoking the Torch mathematical examples to confirm it can perform calculations but when invoking the PyTorch transformers library for semantic analysis, it throws the architectural warning messages.

7

Hi,

The warming will be removed by the PyTorch team.
For now, it can be turned off with the following setting:

import warnings
warnings.filterwarnings("ignore", message=".*Found GPU.*compute capability.*")

You can find more information on the topic below:

Thanks.