How to install Jetpack 4.2 on Jetson TX2 and CUDA10 on Ubuntu 18.04

Recently, the Nvidia has published the Jetpack 4.2. Thanks to the Jetson nano. But, the Nvidia has changed the method for installing the jetpack for Tx2. The new method is SDKManager. By the SDKManager, we can install the cuda10 both for the Host machine and Jetson TX2. During the installation, I have met some problems. I hope to help others, who is going to meet the same problem.

1. The installation of the Jetson TX2

1.1 Download the SDKManager

1.2 Installation for the python on the host machine

python 2.7 and SDK manager might not mention if it is missing

# sudo apt install -y python

1.3 Installation of the SDKManager

# sudo apt install ./sdkmanager-[version].[build#].deb

# sdkmanager

1.4 Install jetpack 4.2 for the tx2 and cuda10 for host machine

1.4.1 choose the hardware

in “Hardware Configuration”, choose the “host machine” and “target hardware – jetson tx2”

1.4.2 host components

choose everything for the host components

1.4.3 Installation of Jetson OS

There are 2 parts. Jetson OS and Jetson SDK components. Only choose the Jetson OS. Go to the next step. In the “Set the device to USB Recovery Mode using Automatic or Manual setup:” , choose “Manual setup”. And fellow every step.

1.4.4 Installation of the Jetson SDK components

Restart the SDKManager, in the “Target components”, only choose the “Jetson SDK component”. Becuase the Jetson TX2 install the “Jetson SDK components” from the internet. Connect the RJ45 with another PC, which is sharing its internet. Then fellow every step.

2. Install the Nvidia Graphic card Driver in the host machine.

Also, we have to install the driver for the host.

# sudo apt-get purge nvidia*

# sudo add-apt-repository ppa:graphics-drivers/ppa

# sudo apt-get update

# sudo apt-get install dkms build-essential linux-headers-generic

# ubuntu-drivers device

The cuda10 will only with Nvidia driver above 410. So we choose the driver above the 410

# sudo ubuntu-drivers autoinstall

# sudo reboot

# nvidia-smi

3. Post installation for the cuda10 on the host

# sudo vim /etc/profile

Add the 2 lines to the end of the file

# source /etc/profile

3.1 Test the result

# cd /usr/local/cuda/samples/1_Utilities/deviceQuery

# sudo make

# ./deviceQuery

3.2 Copy the samples

Copy the samples to the home folder, we can compile the samples and run the samples in the home folder without changing the original samples.

# ~

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.