NVIDIA Jetson Nano: Flash, Boot, and First Task

Flash NVIDIA’s image to an SD card, boot the Jetson Nano with a monitor or headless, then blink an LED and run object detection on the GPU.

Last updated August 3, 2026

20 min read

intermediate difficulty

NVIDIA Jetson Nano: Flash, Boot, and First Task

The Jetson Nano is a small single-board computer with a real GPU, built for running AI at the edge. This tutorial takes it from a blank SD card to a working board: downloading NVIDIA’s image, flashing it, booting either with a monitor or fully headless, and then running a first task - blinking an LED, and running an object-detection model on the GPU.

What You Will Need

Jetson Nano Developer Kit (2GB or 4GB)
microSD card, 32GB or larger, UHS-1 speed or better
Power supply - 5V/4A barrel jack recommended, or 5V/2A micro-USB for light use
A host computer to download and flash the image
Either a monitor + USB keyboard/mouse (desktop setup), or nothing extra (headless setup)
An Ethernet cable or a supported Wi-Fi dongle for network access

On power: AI workloads spike the current draw, and a micro-USB supply browns out under load - the board simply reboots mid-task. For anything beyond first light, use a 5V/4A barrel jack and fit the J48 jumper, which tells the Nano to take power from the barrel instead of USB.

# From another computer on the same network. Find the Nano's IP from your
# router, or run 'ip addr' on the Nano over the serial console first.
ssh embedclub@192.168.1.42

# First login walks through the same Ubuntu setup as the desktop path:
# licence, language, keyboard, timezone, username, password.

# blink.py - blink an LED wired to pin 12 (BOARD numbering).
# Wiring: pin 12 -> 220 ohm resistor -> LED anode (long leg);
#         LED cathode (short leg) -> pin 6 (GND).
import Jetson.GPIO as GPIO
import time

LED = 12
GPIO.setmode(GPIO.BOARD)      # use the physical pin numbers on the header
GPIO.setup(LED, GPIO.OUT)

try:
    while True:
        GPIO.output(LED, GPIO.HIGH)
        time.sleep(1)
        GPIO.output(LED, GPIO.LOW)
        time.sleep(1)
except KeyboardInterrupt:
    pass
finally:
    GPIO.cleanup()            # release the pin cleanly on Ctrl+C

# Run NVIDIA's "Hello AI World" object detection on a live image, using the
# pre-built jetson-inference container so nothing has to be compiled.
sudo docker run --runtime nvidia -it --rm --network host \
  --volume ~/jetson-inference/data:/jetson-inference/data \
  --device /dev/video0 \
  dustynv/jetson-inference:r32.7.1

# Inside the container, detect objects in a sample image:
detectnet images/peds_0.jpg images/out.jpg
# 'out.jpg' now has boxes drawn around every person the model found.

# The 4GB Nano runs out of RAM fast under AI workloads. Add 4GB of swap:
sudo systemctl disable nvzramconfig
sudo fallocate -l 4G /mnt/4GB.swap
sudo chmod 600 /mnt/4GB.swap
sudo mkswap /mnt/4GB.swap
sudo swapon /mnt/4GB.swap
# Make it permanent across reboots:
echo '/mnt/4GB.swap swap swap defaults 0 0' | sudo tee -a /etc/fstab

NVIDIA Jetson Nano: Flash, Boot, and First Task

NVIDIA Jetson Nano: Flash, Boot, and First Task

What You Will Need

Step 1: Download the Image from NVIDIA

Step 2: Flash the Image to the SD Card

Step 3: First Boot

Option A - With a Monitor

Option B - Headless

Step 4: First Task - Blink an LED

Step 5: Run an AI Model on the GPU

Troubleshooting

Where to Go Next