Learn / NVIDIA Jetson / Setup and first boot

First Boot on a Jetson Orin Nano

ReviewedBeginnerFreeAbout 6 min
NANO
In short

First Boot on a Jetson Orin Nano: the short answer

The Jetson Orin Nano Developer Kit boots from a microSD card or an NVMe SSD. The simple route is to download NVIDIA's JetPack SD card image, write it with Balena Etcher or Raspberry Pi Imager's custom image option, and boot. Example: On a Jetson, GPU memory is system memory. Free it by running headless and keep swap on fast storage.

Level
Beginner
Time
About 6 min
Topic
NVIDIA Jetson
You will learn to

Flash JetPack, power the kit correctly and check its health with tegrastats and jtop.

Before this: What a Jetson Is and When to Use One

The Jetson Orin Nano Developer Kit boots from a microSD card or an NVMe SSD. The simple route is to download NVIDIA's JetPack SD card image, write it with Balena Etcher or Raspberry Pi Imager's custom image option, and boot. For NVMe or for updating everything at once, use NVIDIA SDK Manager on an Ubuntu PC. Kits that left the factory with older firmware need a firmware update before they can boot JetPack 6; NVIDIA's Getting Started guide for the Orin Nano walks through it, and it only needs doing once.

Power the kit with its 19 V barrel-jack adapter, not from USB. Connect a DisplayPort monitor, keyboard and mouse for the first boot, accept the licence, create a user, and connect to Wi-Fi or Ethernet. After that you can work over SSH like a Raspberry Pi. Use a fast card (A2 class) or, much better, an NVMe SSD: AI models and Docker containers are large and load slowly from a card.

Two tools show what the board is doing. sudo tegrastats prints CPU and GPU load, memory use, temperatures and power once a second. jtop, installed with sudo pip3 install -U jetson-stats and a reboot, gives the same information as a friendly dashboard and lets you change the power mode. Check these whenever something is slow: a full memory or a hot chip explains most surprises.

Check the software before you begin

A Jetson Orin Nano developer kit runs a Linux system from JetPack. A new kit may come with older firmware, and a newer JetPack release can need a firmware update before it will boot, so read the current setup guide for your kit and flash the matching image. Prepare a good microSD card or an NVMe drive, since a slow or damaged card is a common cause of odd failures.

Use a power supply of the rated size and a monitor, keyboard and mouse for the first boot, or the USB serial console if you prefer a headless setup.

First setup

Follow the on-screen setup for language, user and network. Then update the system, check the power mode, and confirm that the GPU is visible with the system tools. Install your Python packages in a virtual environment so one project does not break another.

Heat starts to matter under load, so keep air flowing around the unit even in testing.

Good habits from day one

Make a backup image of a working card before you experiment. Keep notes of the JetPack version, since libraries such as PyTorch and TensorRT must match it. Shut down properly instead of pulling the power, which can damage the file system.

Common mistakes
  • Flashing a JetPack version without checking the firmware requirement.
  • Using a slow or poor microSD card.
  • Pulling the power instead of shutting down.
Worked example

The Jetson freezes when loading a big model

  1. jtop shows memory at 7.6 of 7.6 GB just before the freeze.
  2. The CPU and GPU share the same 8 GB, and the desktop uses about 1 GB on its own.
  3. Switch off the desktop for headless work with sudo systemctl set-default multi-user.target, and add swap on the NVMe SSD.
  4. Reboot and load the model again: memory peaks around 6 GB and nothing freezes.

Result. On a Jetson, GPU memory is system memory. Free it by running headless and keep swap on fast storage.

Try it

Flash JetPack, complete the first-boot setup and install jtop. Run jtop while a browser plays a video, and note the GPU load, the memory used and the board temperature.

Check yourself

3 quick questions

  1. How should the Orin Nano Developer Kit be powered?

  2. Which tool shows CPU, GPU, memory, temperature and power once a second?

  3. Why prefer an NVMe SSD over a microSD card?

Common questions

What will I learn in "First Boot on a Jetson Orin Nano"?

Flash JetPack, power the kit correctly and check its health with tegrastats and jtop.

How do I the Jetson freezes when loading a big model?

jtop shows memory at 7.6 of 7.6 GB just before the freeze. The CPU and GPU share the same 8 GB, and the desktop uses about 1 GB on its own. Switch off the desktop for headless work with sudo systemctl set-default multi-user.target, and add swap on the NVMe SSD. Reboot and load the model again: memory peaks around 6 GB and nothing freezes. On a Jetson, GPU memory is system memory. Free it by running headless and keep swap on fast storage.

How should the Orin Nano Developer Kit be powered?

With its 19 V barrel-jack adapter. The kit is designed for the supplied 19 V adapter.

Which tool shows CPU, GPU, memory, temperature and power once a second?

tegrastats. sudo tegrastats is built in; jtop gives a friendlier view.

Why prefer an NVMe SSD over a microSD card?

Models and containers are large and load far faster from an SSD. Storage speed matters for loading models and Docker images.