Setting up Common Software on Jetson AGX Orin

TL;DR

In general, we need to install the aarch64 distro of the software, since this device is using the ARM64 ISA.

Install CUDA

CUDA should be included with the Jetpack 6.1 SDK.

To test installation, do

nvcc --version

Install VSCode

When downloading from the official website, select ".deb - Arm64" button.

After download, use the following command to install.

sudo apt install ~/Downloads/code_1.95.1-1730354713.deb

Install MiniForge

Download "Mambaforge-24.9.0-0-Linux-aarch64.sh" from the official release page.

chmod +x ~/Downloads/Mambaforge-24.9.0-0-Linux-aarch64.sh
~/Downloads/Mambaforge-24.9.0-0-Linux-aarch64.sh

Install PyTorch

PyTorch with CUDA support is not available from the normal pip installation method.

Method 1: install from whl file

Older version of PyTorch can be found here.

First, we need to install dependencies for sparse matrix operation support

Follow instructions here.

wget https://developer.download.nvidia.com/compute/cusparselt/0.6.3/local_installers/cusparselt-local-tegra-repo-ubuntu2204-0.6.3_1.0-1_arm64.deb
sudo dpkg -i cusparselt-local-tegra-repo-ubuntu2204-0.6.3_1.0-1_arm64.deb
sudo cp /var/cusparselt-local-tegra-repo-ubuntu2204-0.6.3/cusparselt-*-keyring.gpg /usr/share/keyrings/
sudo apt-get update
sudo apt-get -y install libcusparselt0 libcusparselt-dev

also numpy is recommended

pip install numpy

Download the PyTorch wheel for JetPack 6.1.

After download, do

pip install ~/Downloads/torch-2.5.0a0+872d972e41.nv24.08.17622132-cp310-cp310-linux_aarch64.whl

Method 2: using docker containers

NVIDIA also provided a couple of preconfigured containers that can be installed with the jetson-container command. Following this instruction.

cd ~/Desktop/
git clone https://github.com/dusty-nv/jetson-containers
bash jetson-containers/install.sh

Edit /etc/docker/daemon.json

daemon.json
{
    "runtimes": {
        "nvidia": {
            "path": "nvidia-container-runtime",
            "runtimeArgs": []
        }
    },

    "default-runtime": "nvidia"
}
sudo systemctl restart docker

If docker fails to start with error message "Failed to start Docker Application Container Engine", use this command to print the error message

sudo journalctl -u docker.service

Then, do

jetson-containers run dustynv/pytorch:2.1-r36.2.0

Testing Installation

Test if installation succeed

python
> import torch
> torch.zeros(64).cuda()

Install Jtop

This is a helpful system resource monitor tool, similar to htop and nvidia-smi.

sudo pip3 install -U jetson-stats

After install, a system restart is required.

Last updated

Was this helpful?