Jetson Thor: Containers does not see GPU

Robotics & Edge Computing Jetson & Embedded Systems Jetson Thor
1

I am following the instructions on this page https://docs.nvidia.com/jetson/agx-thor-devkit/user-guide/latest/setup_cuda.html and https://docs.nvidia.com/jetson/agx-thor-devkit/user-guide/latest/setup_docker.html#example-1-run-pytorch-container

Not able the detect GPU on Thor. Output below.

image
image1645×1047 356 KB

4

Hi,

Which container do you use?
Our latest PyTorch container for Thor is nvcr.io/nvidia/pytorch:25.10-py3.

Thanks.

5

I am using container version nvcr.io/nvidia/pytorch:25.08-py3
The issue is it does not detect GPU. I have tried the same in 2 Thor setups purchased from Arrow. Both shows same issue.

I am getting similar issue when I ran CUDA container. The GPU is not being detected.

This is as part of install instrction posted in https://docs.nvidia.com/jetson/agx-thor-devkit/user-guide/latest/setup_docker.html#example-1-run-pytorch-containerhttps://docs.nvidia.com/jetson/agx-thor-devkit/user-guide/latest/setup_docker.html#example-1-run-pytorch-container

6

Hi,

The error shows an insufficient CUDA driver so this might relate to the setup.

How do you set up your system? Do you use JetPack 7.0?
Could you run nvidia-smi outside of the container and share the output with us?

Thanks.

7

Yes, I used Jetpack 7.0.
nvidia-smi and nvidia-smi -q output are attached.

output.txt (10.8 KB)

nvidia-smi-thor
nvidia-smi-thor1322×597 31.3 KB

8

Hi,

We test the command/container mentioned in the doc and it can work as expected in our environment.

$ docker run --rm -it \
    -v "$PWD":/workspace \
    -w /workspace \
    nvcr.io/nvidia/pytorch:25.08-py3

# python3 <<'EOF'
import torch
print("PyTorch version:", torch.__version__)
print("CUDA available:", torch.cuda.is_available())
if torch.cuda.is_available():
    print("GPU name:", torch.cuda.get_device_name(0))
    x = torch.rand(10000, 10000, device="cuda")
    print("Tensor sum:", x.sum().item())
EOF
PyTorch version: 2.8.0a0+34c6371d24.nv25.08
CUDA available: True
GPU name: NVIDIA Thor
Tensor sum: 49997884.0

Could you try it again?
Or could you try to setup the environment with below steps again?

Thanks.