Nvidia contianer l4t-pytorch:r35.1.0-pth1.11-py3 running on Jetson AGX Orin, pytorch can not using gpu

[sa515125]

Hello，everyone.
I am using Jetson AGX Orin for my project. And the env of the host is likes below:

I deployed a pytorch image provided by Nvidia(NVIDIA L4T PyTorch | NVIDIA NGC) on the machine, but pytorch reported that it cannot detect GPUs. Container Info like below:

The abnormal behavior is as follows：

However, I am able to use GPU smoothly using the Jetpack6 image, but my project currently requires the use of Jetpack5 image.

Is there a restriction that requires the host and image to use the same jetpack version? Or AGX Orin still requires the use of JetPack6 software version?

[AastaLLL]

Hi,

Since PyTorch is built with a specific CUDA version, please use the container that is aligned to your JetPack environment.

For example, nvcr.io/nvidia/l4t-pytorch:r35.2.1-pth2.0-py3 for JetPack 5.
nvcr.io/nvidia/pytorch:24.08-py3-igpu for JetPack 6.

Thanks.