Python 3.13 GPU on Jetson Nano with Headless Mac M3

I'll set up my Jetson Nano as a headless device, connecting it to my MacBook via USB-C to Ethernet. The Mac will be on Wi-Fi, sharing its internet with Jetson while I access it over SSH, ensuring Jetson operates without Wi-Fi.

1. Product

• Jetson Orin nano (8GB)

• USB 3.1 c-type to RJ45

• M3 Macbook

2. Setup

1. Physical Connection

• Connect the Mac and Jetson using a USB-C to LAN adapter and an Ethernet cable.

2. Mac Configuration

• System Preferences > Network > Ethernet(LAN):
  • Select "Configure Using DHCP" to automatically assign an IP (typically 192.168.2.1).
• System Preferences > Sharing:
  • Enable "Internet Sharing".
  • Share from: Wi-Fi, To computers using: Ethernet(LAN).

3. Jetson Configuration

• Install and enable SSH server:

$ sudo apt update
$ sudo apt install openssh-server
$ sudo systemctl enable ssh
$ sudo systemctl start ssh

• Disable Wi-Fi (use Mac connection only):

$ nmcli radio wifi off

• Set a static IP for the Ethernet interface:

$ sudo nmcli connection modify "Wired connection 1" ipv4.method manual ipv4.addresses "192168.2.2/24" ipv4.gateway "192.168.2.1" ipv4.dns "8.8.8.8"
$ sudo nmcli connection up "Wired connection 1"

4. Verify Connection

• SSH into Jetson from Mac:

$ ssh jetson@192.168.2.2

• Check internet connectivity from Jetson:

$ ping google.com

4. Cuda12 & Pytorch2.*

$ sudo apt install -y python3.12
$ python3 --version

# cuda
$ sudo apt install -y ubuntu-drivers-common
$ sudo ubuntu-drivers autoinstall
$ sudo apt-get install -y cuda-toolkit-12-6

$ echo 'export PATH=/usr/local/cuda-12.6/bin:$PATH' >> ~/.bashrc
$ echo 'export LD_LIBRARY_PATH=/usr/local/cuda-12.6/lib64:$LD_LIBRARY_PATH' >> ~/.bashrc
$ source ~/.bashrc

$ nvidia-smi # check

# pytorch 2.x
$ python3.12 -m venv .venv
$ source .venv/bin/activate
$ pip install torch --index-url https://download.pytorch.org/whl/cu126
$ pip install torchvision torchaudio
$ python3 -c "import torch; print('PyTorch version:', torch.__version__); print('Device:', torch.device('cuda' if torch.has_cuda else 'cpu'))"