Jetson Xavier Setup and Installation Instructions

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

  1. Create an Nvidia Developer account.
  2. Download and Install Jetpack 4.2 SDK Manager on your Host PC.
  3. Connect Xavier to Host using the USB-C cable (use front-port).
  4. Run the SDK Manager on the Host and follow the installation instructions.
  5. 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.
  6. Once OS Setup is complete on Xavier, finish the installation process on the Host.
  7. 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:

Shell
1
2
3
4
5
6
7
8
9
10
11
12
13
sudo apt install libhdf5-serial-dev hdf5-tools
sudo apt install python3-pip
pip3 install -U pip
cd /usr/bin/
sudo chmod 777 pip3
 
#Open pip3 in a text-editor and change:
#from pip import main => from pip import __main__
#sys.exit(main()) => sys.exit(__main__._main())
 
sudo apt install zlib1g-dev zip libjpeg8-dev libhdf5-dev
sudo pip3 install -U numpy grpcio absl-py py-cpuinfo psutil portpicker six mock requests gast h5py astor termcolor
pip3 install --pre --extra-index-url https://developer.download.nvidia.com/compute/redist/jp/v42 tensorflow-gpu

Test installation by running:

Shell
1
2
python3
>> import tensorflow

This command should run without error.

Download and move the models folder into tensorflow.

1
2
3
4
5
6
7
8
sudo apt install protobuf-compiler python-pil python-lxml python-tk
pip3 install --user Cython
pip3 install --user contextlib2
pip3 install --user jupyter
sudo apt install python3-matplotlib
cd tensorflow/models/research
protoc object_detection/protos/*.proto --python_out=.
export PYTHONPATH=$PYTHONPATH:`pwd`:`pwd`/slim

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

Shell
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
#Move scripts file to xavier
sudo scp <<HostFilePath>> username@192.168.55.1:<<XavierFilePath>>
 
#Get fan speed
cat /sys/kernel/debug/regulator/vdd-fan/regulator/pwm-fan-vdd-fan/curr_pwm
 
#Set fan speed to max (255)
echo 255 | sudo tee /sys/kernel/debug/regulator/vdd-fan/regulator/pwm-fan-vdd-fan/target_pwm
 
#Max performance
sudo /usr/bin/jetson_clocks
 
#Power Mode, 2 is default, 0 is MAX.
Check: sudo nvpmodel -q
Set: sudo nvpmodel -m #

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.


How To Select Right Deep Learning Model For Object Detection Applications