This is a custom image I have created for the Jetson Nano. ( Download link at the bottom. )
Built from the ground up using debootstrap, this image is based on a minimal Xubuntu-core / Xfce4 installation for a snappy desktop experience and a lighter weight alternative to the L4T Ubuntu image. This image is only 3.5 Gigabytes.
INSTRUCTIONS
(Please note: Your device must be rebooted at least once before all the drivers will be loaded correctly. The first boot is a good time to create any user accounts.)
Default Username: jetson Default Password: 12345
Steps to create a new user and delete the default
Download the archive and decompress it. (image size= 3.5 GB)
Burn the .img with Belena Etcher or another tool
Pop it in and boot the nano!
Login using Username: jetson Password: 12345
Open a term and sudo adduser somenewadmin
sudo usermod -aG sudo somenewadmin
sudo adduser somenewadmin adm
sudo adduser somenewadmin plugdev
exit (and log in as somenewadmin)
Open terminal again and sudo deluser --remove-home jetson
Notes
The image contains the following:
The L4T-Jetson-Driver-Package-R32.1 (Everything thats normally installed with this package is included, you can check nvidias package manifest if you donāt already know what it is)
Xubuntu-core (basically everything thatās included in a typical minimal xubuntu/xfce4 installation)
Chromium-browser
VLC Media Player
Compton ( Compositor program to make everything silky smooth )
A couple GTK themes and a wallpapers to make things look pretty.
And thatās pretty much it! , Iāll attach a package manifest for everyone to take a look at.
Compton is set to autostart and the configuration file is added to the .config folder in the $HOME directory ā¦ feel free to play around with the settings but what I have feels pretty nice.
If there is anything anyone thinks I can do to improve this image let me know! if there is interest iād like to create an image that contains everything the L4T image does like all the development tools , The Media API package , Cuda repositories and whatnot but as it stands now people can just use the SDK tool to install what they need.
I probably sound pedantic (and I can be) or like a broken record, but I feel it needs to be said for security sake.
Adding a root password isnāt safe. Thatās one reason the sudo group exists (so the root password can be disabled entirely as a security measure).
You should be able to add new users and delete the old with:
(directly after first boot as user Jetson)
sudo adduser somenewadmin
sudo adduser somenewuser
sudo usermod -aG sudo somenewadmin
exit (and log in as somenewadmin)
sudo deluser --remove-home jetson
That should be it after first boot. Use the regular user unless you need to modify the system. Itās wise to work this way on Windows as well.
I attempted to set up the ubiquity installer along with the oem-config settings so the installation would function the same as the nvidia image however I ran into some issues so i just reverted to using a default account.
Thatās correct scottkvps23 you need to install the sdkmanager on an x86 computer that is running Ubuntu 18.04 or you can install it in a virtual machine.
Then with the nano booted up and logged in you attach a micro usb cable from the nano to the x86 computer running the sdkmanager and it will allow you to install the extra packages.
There are also .deb packages in the Sdk manager dowbload path. Ones labeled āarm64ā are for the nano itself.
Also, a MicroUSB cable is not necessary to use SDK manager to provision a machine. Turns out if you have ssh open and use the up and credentials of the Nano that works too.
I donāt think SDK manager will connect to a custom image without ssh running. Itās existing connection relies on a network interface over MicroUSB and that is started by a service Nvidia added.
I absolutely hate now Nvidia set this up too. Why canāt apt just work? I have no idea. Why is machine A required to install something on machine b? You got me. Please, Nvidia, retire SDKM.
On your Ubuntu host PC, after running sdkm and installing everything, if you are very observent you may spot a sneaky option for download location. The dpkg files get put there.
Sorry I didnāt read carefully enough. I donāt think you have a choice unless Nvidia had a sdkm port for fedora/suse. There is also a tarball with the debs to install on nano but I canāt recall the download url.
As a side note, this custom image has all the same functionality as the sample rootfs available from nvidia so if you can do something with that image you should be able to do it with this one as well.
I an not 100% certain but it would make sense they used ssh because if I change the IP to that of a real interface with ssh running is still works. Iām sure itās not ssh for the flashing bit, bit for the install bit it appear to be.
Unfortunately there are known issues with SDKM in a VM. It must be run on bare metal because of the way it handles USB.
Did you try flashing and the install both? I donāt think that will work. It might, but itās a configuration Nvidia doesnāt support and explicitly warns against.
In reference to the tarballs I mentioned earlier, the .deb packages are in one of these downloads as well but I canāt recall which. When I get off mobile I will check which:
edit: I canāt find them. I may be mistaken about them being in a tarball, or I may be failing at searching.
Sorry youāre correct, I have not attempted to flash the nano via the sdkmanager on a vm.
However I have not had any issues installing the CUDA,AI,ComputerVision and Multimedia components via a usb cable connected to a host machine running an ubuntu 18.04 virtualbox image.
Youāre right that nvidia should host the components in an online repo ā¦it would make things a lot easier.
they should also release a package manifest for all the development tools they are pre-installing in their L4T image so ppl know whats included.
I was only able to get that to work over ssh. I plugged my nano into my workstation and all hid devices stopped responding. The particular workstation has what I suspect are usb power issues. It shouldnāt happen, especially since the nano is powered through the barrel jack, but it does. These are the sorts of edge cases Nvidia didnāt think about but happen all the time.
I think Nvidia is used to embedded developers who are willing to install their software on their host and all that it implies. The problem is when you advertise something as āUbuntuā people are going to expect it to be just that, not Nvidia customized green-everywhere Ubuntu with some software you have to install on your (Ubuntu on bare metal only) host for reasons nobody can quite explain.
The second is what most people coming from debian based Linux are used to. Update the system, apt-get install some stuff, add a gpg key, add a repo, apt-get update, and apt-get install the-thing.
Itās what people have already done many times and are hopefully used to. It just works, and will continue to work, and can auto-update, and can work in a container, andā¦
