Getting Started with Jetson Orin Nano (Development Kit) for Real

This is the official NVidia getting started guide. But There are so many things that are conveniently omitted from that guide.

To get Jetson Orin Nano (Development Kit) running, you have to buy:

* An SD card

* An DisplayPort to HDMI Adapter (4K DP to HDMI Adapter) (see discussion here)

What is called "flashing" the SD card isn't really what we all mean by flashing...

> If using JetPack 6.x SD Card image for the first time, you will need to update the QSPI bootloaders by installing JetPack 6 on your SD Card using SDK Manager, which will update the QSPI bootloaders as well. Please note that this is a one time requirement only. Once the QSPI bootloaders are updated, you can use JetPack 6.x SD card images for any future releases.

> Whether you have the SD card image or not, you will always need an Ubuntu PC to update the QSPI memory on the module, so the bootloader is compatible with JetPack 6, which is not possible with a Windows PC.

- Installing Jetpack 6.0 on New Orin Nano with SD Card Image Method

The guide with the actual way more involved process can be found here. You have to download and install the SDK Manager. See here.

To do this, there are some Docker containers which might work but highly likely they won't work. After you install Docker to work with WSL and all that jazz, you might end up with a screen like this:

What you really need is an Ubuntu 22.04 Windows PC at this point and time. As soon as you turn your old laptop to a dual-boot Ubuntu, you are about to start making progress on "flashing".

Flashing doesn't mean using an SD card writer to write an image (e.g. with balenaEtcher). You will have to run the manager and connect the Jetson Orin Nano to the PC with a USB-C to -whatever your old laptop has connector, and then you have to put Jetson Orin Nano in Force Recovery Mode. To do that you need to have the SD card in the Jetson Orin Nano and put a jumper while you boot the device on pins 9 and of the jumper header that is (barely visible) under the GPU. J14 is NOT the 40-position easily accessible jumper you see on the right of the device. After you do that, the SDK Manager will download software for about an hour and do various operations. At some point, it will start to flash an image to Jetson Orin Nano. Then it will try to connect with SSH over the USB and install additional software. This might not work at first. Be sure to fill-in your passwork correctly and be patient and wait for Jetson Orin Nano to (quietly) reboot while the SDK Manager retries to connect. This might take a couple of stressful minutes. Then there will be more transfers for a while.

At the end of this process the Jetson Orin Nano will be successfully flushed.

It will have some software in it, but it's not the best installation I've seen. You will need to update Python venv with apt get. You will also need a browser like chromium-browser. Do not use the original Tensorflow releases - you might get errors like "ERROR: Failed building wheel for h5py" (e.g. here) or not getting to see any GPU (e.g. here). Jetson Orin Nano works easier with the custom NVidia builds for Orin - something like this:

$ sudo pip3 install --extra-index-url <url> tensorflow==<VERSION>

You also have to install prerequisites. The URL/tensorflow on the page above might not be the latest available. Here's an example of installation with a more recent version (v60dp), today.

Note also that nvidia-smi is present but won't work because the GPU isn't connected through PCI. People suggest to use the "deviceQuery" example, but don't be surprised if you don't find it on Jetson Orin Nano. Maybe htop or jtop will give better results as described here.