JetPack 6 Key Feature: Upgradable Compute Stack

From RidgeRun Developer Wiki



Previous: JetPack 6 Key Features/Bring Your Own Kernel Index Next: JetPack 6 Key Features/New Linux Distributions






In this section, you will find comprehensive guidance on maintaining and upgrading the Compute Stack within JetPack 6. It details the importance of keeping the Compute Stack current, and provides step-by-step instructions for upgrading various components.


The Compute Stack in JetPack 6 serves as the backbone for efficient and high-performance computing, integrating critical components such as libraries, and frameworks required for various applications. Keeping the Compute Stack up-to-date ensures optimal performance, enhanced security, and compatibility with the latest technologies. Upgrading the Compute Stack allows users to leverage new features, improvements, and bug fixes, facilitating smoother and more efficient workflows.

With the introduction of an upgradable compute stack supported from the JetPack 6.0 production release, developers can now upgrade to the CUDA-X compute stack from the latest JetPack releases on previous JetPack versions without upgrading the Jetson Linux BSP. This feature supports running the Compute stack on two prior JetPack major dot releases. The Figure 1 illustrates an example.


Diagram showing compute stack
Figure 1. Upgradable Compute stack example on JetPack 6.0. Extracted from link


Since the only two releases are JetPack 6.0 developer preview and production release, this feature can currently only be leveraged to use the JetPack 6 production release compute stack within the developer preview. To use CUDA, CUDNN, TensorRT, DLA, and VPI from the JetPack 6.0 production release on the JetPack 6.0 Developer Preview release, follow the steps below on a Jetson Orin with JetPack 6.0 DP:

1. Add R36.3 repo.

1.1. Open the apt source configuration file in the text editor of your choice, for example:

sudo vi /etc/apt/sources.list.d/nvidia-l4t-apt-source.list
sudo emacs /etc/apt/sources.list.d/nvidia-l4t-apt-source.list
sudo nano /etc/apt/sources.list.d/nvidia-l4t-apt-source.list

1.2. Change the repository name and download URL in the deb commands by the following:

deb https://repo.download.nvidia.com/jetson/common r36.3 main
deb https://repo.download.nvidia.com/jetson/t234 r36.3 main

1.3. Save and close the source configuration file.

1.4. Enter the following commands:

sudo apt-get update
sudo apt-get dist-upgrade

If apt prompts you to choose an option to fix local modified configuration file, reply Y for yes (to use the NVIDIA updated version of the file).

If you are upgrading from release 34, install the following new packages manually:

sudo apt-get install nvidia-l4t-openwfd \
                     nvidia-l4t-vulkan-sc-dev \
                     nvidia-l4t-vulkan-sc-samples \
                     nvidia-l4t-vulkan-sc-sdk \
                     nvidia-l4t-vulkan-sc \
                     nvidia-l4t-jetsonpower-gui-tools

1.5. When the upgrade is finished, reboot the Jetson device.

2. Remove JetPack meta packages if installed:

sudo apt-get remove nvidia-jetpack \
                    nvidia-jetpack-runtime \
                    nvidia-jetpack-dev

3. Install DLA compiler package:

sudo apt-get install nvidia-l4t-dla-compiler

4. Install JetPack runtime packages:

sudo apt-get install nvidia-cuda \
                     nvidia-tensorrt \
                     nvidia-cudnn8 \
                     nvidia-cupva \
                     nvidia-opencv \
                     nvidia-container \
                     nvidia-vpi

5. (Optional) Install JetPack development packages if needed:

sudo apt-get install nvidia-nsight-sys \
                     nvidia-cuda-dev \
                     nvidia-tensorrt-dev \
                     nvidia-vpi-dev \
                     nvidia-opencv-dev \
                     nvidia-cudnn8-dev

6. Remove the R36.3 repo to avoid installing nvidia-l4t-gstreamer accidentally later. Do not use apt-get install nvidia-jetpack or apt-get upgrade because that will upgrade nvidia-l4t-gstreamer, too.



Previous: JetPack 6 Key Features/Bring Your Own Kernel Index Next: JetPack 6 Key Features/New Linux Distributions