Jetson Orin Nano Configuration

System Requirements

Host Machine (for SDK Manager)

  • OS: Ubuntu 22.04 LTS (recommended for SDK Manager)

  • Architecture: x86_64

  • Internet connection: Required

  • RAM: Minimum 8 GB (16 GB recommended)

  • Storage: Minimum 43 GB free space (64 GB+ recommended)

  • USB cable for Jetson recovery mode

Target Device

  • NVIDIA Jetson Orin Nano Developer Kit

  • microSD card or NVMe SSD

  • Power supply

Installation

Prepare Host Machine

  1. Boot a PC/laptop with Ubuntu 22.04 LTS.

  2. Install NVIDIA SDK Manager.

  3. Connect the Jetson Orin Nano using USB.

  4. Put the Jetson into Force Recovery Mode.

Install SDK Manager

Download and install SDK Manager:

wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/x86_64/cuda-keyring_1.1-1_all.deb
sudo dpkg -i cuda-keyring_1.1-1_all.deb
sudo apt update
sudo apt install sdkmanager# Jetson Nano Configuration

Launch:

sdkmanager

Flash JetPack SDK

In SDK Manager:

  1. Log in with NVIDIA account

  2. Select:

Product Category:
Jetson

Hardware:
Jetson Orin Nano Developer Kit

JetPack:
JetPack 7.x

  1. Select Target Storage:

  • MicroSD Card

  • NVMe SSD

  1. Start installing

SDK Manager will install:

  • Jetson Linux (L4T)

  • Ubuntu 24.04 target system

  • CUDA

  • cuDNN

  • TensorRT

  • NVIDIA drivers

  • Multimedia libraries

Verify Installation

On the Jetson:

Check JetPack:

cat /etc/nv_tegra_release

Check CUDA:

nvcc --version

Check GPU:

nvidia-smi

Environment Configuration

Target stack:

Jetson Orin Nano
│
├── JetPack 7
│   ├── NVIDIA Driver
│   ├── CUDA
│   ├── cuDNN
│   └── TensorRT
│
├── Docker + NVIDIA Container Runtime
│
└── YOLO Environment
    ├── PyTorch
    ├── Ultralytics
    ├── OpenCV
    └── Your SmartDB AI application

First boot

On the jetson:

sudo apt update
sudo apt upgrade -y

Install tools:

sudo apt install -y \
git \
curl \
wget \
vim \
htop \
python3-pip \
python3-venv \
build-essential

Install Jetson Monitoring tools

sudo pip3 install jetson-stats

Reboot:

sudo reboot

Run:

jtop # like htop but for jetson

Install NVIDIA container runtime

Docker is recommended for your YOLO environment.

Install Docker:

sudo apt install -y docker.io

Enable:

sudo systemctl enable docker
sudo systemctl start docker

Add yourself:

sudo usermod -aG docker $USER

Logout/login.

Install NVIDIA container runtime:

sudo apt install -y nvidia-container

Test:

docker info

Install NVIDIA container toolkit

For Jetson:

sudo apt install nvidia-container

Restart Docker:

sudo systemctl restart docker

Test GPU access:

docker info | grep NVIDIA

Create YOLO workspace

mkdir ~/smartdb_ai
cd ~/smartdb_ai

Structure:

smartdb_ai/
├── models/
│   └── best.pt
├── src/
│   ├── detector.py
│   └── camera.py
├── datasets/
└── Dockerfile

Use an NVIDIA L4T PyTorch container.

Example:

docker run --runtime nvidia \
--network host \
--volume ~/smartdb_ai:/workspace \
-it nvcr.io/nvidia/l4t-pytorch:r36.4.0

Inside container:

pip install ultralytics
pip install opencv-python

Test:

import torch

print(torch.cuda.is_available())

Expected:

True

Test YOLO

Inside container:

yolo predict \
model=models/best.pt \
source=0

or Python:

from ultralytics import YOLO

model = YOLO("models/best.pt")

results = model("test.jpg")

for r in results:
    print(r.boxes)

Optimize for Jetson

For deployment, don’t run .pt directly.

Convert:

from ultralytics import YOLO

model = YOLO("best.pt")

model.export(
    format="engine",
    half=True
)

Output:

best.engine

Then inference uses TensorRT:

Camera
  ↓
TensorRT YOLO
  ↓
Detection
  ↓
SmartDB logic

Camera setup

USB camera:

sudo apt install v4l-utils

Check:

v4l2-ctl --list-devices