I imagine @dusty_nv will reply, but I will add something first just to clarify some parts which might be confusing.
First, documentation for the Nano itself will come with the particular L4T release. L4T (“Linux for Tegra”) is just Ubuntu plus NVIDIA drivers. You can find your L4T release with “head -n 1 /etc/nv_tegra_release”. Then you can go to that release here:
https://developer.nvidia.com/linux-tegra
Nanos have an SD card for the root filesystem (the rootfs…also know by its partition label “APP”). No Jetson has a BIOS, so the rest of starting a Jetson depends on software which does the equivalent of this, plus bootloader. For SD card models this is on QSPI memory of the module itself, while on eMMC models this is in partitions of the eMMC. You can flash the Nano to update its QSPI memory. This is sometimes needed if you use a newer release of SD card in order for have compatibility between SD card software and QSPI (think BIOS and bootloader) software.
One flashes QSPI of the Nano with JetPack/SDK Manager. JetPack is a GUI front end to the flash software, while SDKM is a smart network layer on top of JetPack. If you pick a JetPack release, then this is tied to the L4T release. Knowing one’s release means knowing the other’s release. You could look up JetPack/SDKM releases here, but note the L4T page tells you which hardware is compatible (JetPack/SDKM is compatible with more varied hardware, but not necessarily with the Nano):
https://developer.nvidia.com/embedded/jetpack-archive
Basically, if you go to one, you’ll end up with the same software that you’d get if you had went to the other. This and documentation.
If your SD card does not work, then the first thing you’d do is to flash the QSPI of the module using JetPack/SDKM. This requires an Ubuntu host PC. For Nano (not Orin, just plain older Nano) you can use Ubuntu 18.04 or 16.04 for the host PC. Command line flash is much more flexibile. If you use a VM, then this tends to be a problem since VMs need configuration and this is out of the realm of the flash software. You’d need to know how to force the USB to always remain connected to the VM (during a flash the USB disconnects and reconnects).
When in recovery mode (which in itself does not change a Jetson) the Jetson becomes a custom USB device understood by the custom driver (and the actual flash software run by the JetPack GUI is appropriately named the “driver package”). This runs only on Linux. Jetsons don’t have a BIOS, so they cannot self-flash. Thus the host PC requirement, and recovery mode. If your SD card is working ok, then no need to flash. A lot of SD card models of Jetsons though really should have the QSPI flashed at least once when they first arrive.
JetPack/SDKM would normally flash a Jetson, and then the Jetson would automatically reboot. It is upon first boot one sets up an account login, and then JetPack uses that login to install optional software. This includes CUDA and some other options. All of this goes through a fully booted Jetson over ssh. If you want to explore optional software which is compatible with the Nano, then get the L4T flash software which is a match to your L4T version (again, see “head -n 1 /etc/nv_tegra_release” to find that). Start sdkmanager as a regular user. You can use a recent JetPack and still see older releases if you start it on command line like this:
sdkmanager --archivedversions
Then uncheck flash operations. By doing this you can go straight to the optional software part. You’ll see checkboxes for various optional software, although you might need more space.
You should also know that Jetsons have an integrated GPU (iGPU) which is directly wired to the memory controller. Most people have a discrete GPU (dGPU) which is going through the PCIe bus. The software and drivers for a dGPU differ from that of the iGPU. The drivers supplied from anything other than JetPack are normally incompatible and will probably break things when mixing a dGPU driver with an iGPU driver. The desktop PC software “nvidia-smi” operates through the PCIe bus, and so it is incompatible with a Jetson (and you won’t find any PCIe-dependent GPU software on a Jetson).
The release of CUDA and much of the AI software (if it talks to the GPU) is tied to the CUDA 10 release on any older Jetson (such as the Nano). You would have to purchase newer hardware to get a newer release, e.g., Xavier can use CUDA 11, and Orin can use CUDA 11 or 12 (the latter via a docker container). Software which is based on a PCIe dGPU will not work. Thus it might be important to install a JetPack 4.x, start it with “sdkmanager --archivedversions”, pick your release, uncheck flash, and see what software shows up for install. Much (not all) of what you want would be from this with a fully booted Jetson.