Start writing In AI/ML, data and computing power are resources that go hand-in-hand. So, as our datasets and databases have grown over the years, so too has the demand for computing power. While it is possible to make use of cloud computing services such as those offered by Microsoft, IBM, Google etc. such an approach can be costly and at times cumbersome. In the last year Nvidia has introduced its Xavier computer as a powerful and efficient solution for developers seeking to prototype their ideas.
The Xavier is an AI computer that delivers the performance of a high-end GPU workstation in an embedded module that operates at 30W or less. Containing a 512-core Volta GPU, 8-core 64-bit CPU, and a 16GB Memory, it’s hard to believe that all of that power can fit in the palm of your hand!
So, you can expect that this is an exciting new piece of technology for all the AI Scientists, ML Engineers, and Data Analysts around the world that are bonded together by our shared need computing power and have ever felt limited by the amount of computing power available to them.
This post will hopefully help anyone who has obtained a Xavier and would like some guidance on how set it up, install a few packages, and configure device behaviour.
Note: Make sure you have all necessary components including the power cord, USB-C cable, a keyboard, mouse, ethernet connection, hdmi cable and monitor for Xavier.
SDK Installation
- Create an Nvidia Developer account.
- Download and Install Jetpack 4.2 SDK Manager on your Host PC.
- Connect Xavier to Host using the USB-C cable (use front-port).
- Run the SDK Manager on the Host and follow the installation instructions.
- When flash is at 50%, you will be prompted to connect Xavier to a monitor with an HDMI cable and follow the setup instructions on the Xavier.
- Once OS Setup is complete on Xavier, finish the installation process on the Host.
- Once the SDK Manager has successfully “flashed” the Xavier, you can disconnect the USB-C connection between Xavier and Host.
Tensorflow Libraries and Packages
Now you can work on the Xavier like a normal PC. In the Xavier terminal:
Test installation by running:
This command should run without error.
Download and move the models folder into tensorflow.
At this point, the TensorFlow framework and several ML libraries have been installed and should be ready for you to experiment with. If everything is installed correctly, the following command should run successfully:
python3 object_detection/builders/model_builder_test.py
Note: Currently, this installation process using pip takes a really long time, don’t think that the installation is stuck, just wait. These instructions were taken from here.
Note: If Xavier cannot connect to the internet through an ethernet connection, use ifconfig to find the device’s MAC address, and make sure that the address is not being blocked on your network.
Note: You should be able to ssh into the device with the USB-C cable at <<username>>@<<IP-Adress>>. The factory IP is 192.168.55.1 but that may change post setup.
System/Config Scripts for Xavier
Find out more about Xavier power modes here.
Remaining Concepts/Issues
- The Xavier only possesses 32G of storage space. In order to maintain the integrity of the device/drivers/libraries it is advisable to use docker. Docker images and containers quickly take up storage space so it is recommended that an SSD is added to the rig.
- Have not yet studied or tested the Xavier’s special features that allow it to accelerate training speeds.
- A suitable USB-powered camera needs to be found in order to carry out live-object detection programs.