CUDA Setup#

JetPack includes the NVIDIA driver stack and CUDA-enabled software support for Jetson Orin Nano Developer Kit. You can use CUDA from containers or install JetPack components directly on the host system.

CUDA Setup Options#

CUDA Setup Flow CUDA Setup Flow

Container-Based CUDA#

Container-based development is recommended when you want a reproducible userspace with prebuilt NVIDIA AI and CUDA packages.

  1. Complete Docker Setup.

  2. Pull a Jetson-compatible CUDA or framework container from NGC.

  3. Run the container with GPU access.

Example:

docker run --rm -it nvcr.io/nvidia/cuda:13.0.0-devel-ubuntu24.04

Native CUDA Packages#

Use native packages if your project needs CUDA tools installed directly on the Jetson Linux root file system.

Install or update JetPack components from the JetPack package repositories:

sudo apt update
sudo apt install nvidia-jetpack

See JetPack SDK Setup for JetPack component guidance.

Verify CUDA#

Confirm CUDA tools and GPU access using a CUDA sample, a framework container, or your project workload.

For a Python framework test inside a PyTorch container:

python3 <<'EOF'
import torch
print("CUDA available:", torch.cuda.is_available())
if torch.cuda.is_available():
    print("GPU name:", torch.cuda.get_device_name(0))
EOF