Connect NVIDIA Jetson AGX Xavier
The NVIDIA Jetson AGX Xavier is a developer board based on an 8-core ARM CPU including a powerful 512-core Volta GPU that enables the device to cope with complex AI tasks.
Last updated
The NVIDIA Jetson AGX Xavier is a developer board based on an 8-core ARM CPU including a powerful 512-core Volta GPU that enables the device to cope with complex AI tasks.
Last updated
The setup of an NVIDIA Jetson is not as straightforward as the setup for a Raspberry Pi. The setup includes several manual steps to be completed before you can successfully connect the device to the Record Evolution Platform.
The NVIDIA-specific setup requires a host (laptop or desktop machine) running Ubuntu 18.04 or Ubuntu 16.04. To install one of these, you first have to download the corresponding iso-image for Ubuntu 18.04(or Ubuntu 16.04) and write it to a USB flash memory or an SD card. This can be conveniently done with the Record Evolution Reflasher.
After successfully writing the iso-image to the flash memory, insert the flash memory into your host machine, restart the machine, and enter the boot menu to boot from the flash memory.
Just follow along the steps during the initial setup process of Ubuntu.
After completing the initial setup and rebooting in the OS, make sure you have the qemu-user-static
, any python2.x
package and an appropriate soft link /usr/bin/python -> /usr/bin/python2.x
to the python2.x binary installed. If you do not have these, open up a terminal and execute:
Given that you have successfully installed Ubuntu 18.04/16.04 on a host machine, you can now start to connect the hardware in the following way:
Do not connect the power plug before everything else is connected!
First, we have to connect the Jetson board with the host machine by means of an USB-C-to-USB-A/C cable (depending on your host machine).
Then we connect the HDMI output of the board to an external screen to able to complete the graphical user setup later.
Finally, we connect an USB keyboard to fill the forms of the graphical setup.
After having connected everything as displayed on the image above, we connect the power plug as well (without pushing the power button of the board!).
Now, we put in the board into so-called FORCE RECOVERY MODE which is done by pushing and holding the three buttons on the side of the board (see the picture below) in the following order:
Push and keep holding the force recovery button (center button).
Push and hold and the power button (left button).
Release the power button.
Release the force recovery button.
This puts the NVIDIA board into force recovery mode. To confirm that, please open up a terminal on your Ubuntu host machine and type:
The expected output is:
If you do not see this result but no output at all, go back to trying to put the board into force recovery mode.
Now, we have to prepare the root file system for the NVIDIA board on our host system.
To this end, download the required latest drivers e.g. the L4T 32.6.1 Release archive and the corresponding root filesystem, e.g. Sample Root Filesystem 32.6.1 to the host machine. Start by extracting the driver archive by:
This produces the directory Linux_for_Tegra
in the same folder. Enter the folder Linux_for_Tegra/rootfs/
and extract the root filesystem into it by:
Here it is ESSENTIAL to do this as root with the flags xpf
in order to preserve the permission and ownership in the root filesystem.
Next, we prepare some binaries by:
After successful completion of this step, we can finally flash the internal board memory (while making sure that the device is still in FORCE RECOVERY MODE):
Depending on your host machine, this may take up to about five minutes to complete.
After the script is successfully finished, the NVIDIA board will automatically reboot and start to show the initial user setup on the connected HDMI screen.
After finishing the basic board setup and completing the initial user setup, make sure to connect the board to a working Ethernet connection.
Now reboot it and log in either directly via the graphical user interface and keyboard or via ssh on the local network.
Proceed with the following steps in order to connect your NVIDIA device to the Record Evolution Platform:
Make sure the system is up-to-date by executing sudo apt-get update && sudo apt-get upgrade
Download the device setup shell script wget https://storage.googleapis.com/reswarmos/nvidia/reswarmify-nvidia.sh
Make sure the script is executable: chmod +x reswarmify-nvidia.sh
Copy your confidential .reswarm
configuration file to the device: scp <your-reswarm-device-config>.reswarm <user-name>@<local-ip-of-board>:
Execute the (idempotent) reswarm script to transform the board into a Record Evolution device: sudo ./reswarmify-nvidia.sh <your-reswarm-device-config>.reswarm
Now log in to your Record Evolution account and check for the connected device. That's it!