What Is NVIDIA JetPack?

NVIDIA JetPack is the software development kit for the Jetson family of edge AI modules. It bundles the operating system, NVIDIA Jetson Linux, with the libraries an edge AI application needs, so a Jetson module can run GPU-accelerated inference rather than arriving as a bare board.

JetPack is versioned, and each version supports a particular set of modules. That makes the version as important as the contents: it fixes the Ubuntu release, the Linux kernel, the CUDA version and the AI library versions your application is built against. Our full guide to NVIDIA JetPack and the Jetson software stack covers each layer in depth; this page focuses on what is in the kit and how the versions line up.

A JetPack release has three main parts.

The table summarizes the main JetPack lines and the modules each supports. Point releases within a line update the libraries and fix bugs without changing the module list.

The original Jetson Nano and TX2 stop at JetPack 4.6, and the Xavier modules at JetPack 5.1. JetPack 6.2 added the Super power modes for Orin Nano and Orin NX. Always confirm the exact module list in NVIDIA’s release notes before choosing a version.

There are three quick ways to find which release a Jetson is running. The file /etc/nv_tegra_release shows the Jetson Linux release, such as R36 revision 4, which maps to a JetPack version. On JetPack 5 and later, apt show nvidia-jetpack reports the JetPack package version directly. The community jtop tool shows JetPack, Jetson Linux and the library versions on one screen.

Record the version, and the carrier board BSP revision, in your build documentation. When a system fails in the field, knowing the exact JetPack, CUDA and TensorRT versions is often the quickest route to the cause.

For a product, the newest JetPack is not automatically the right one. The version has to support the module, the carrier board vendor has to provide a board support package for it, and your application and its dependencies have to be validated on it. Our explainer on board support packages covers why the carrier vendor’s BSP matters.

Major versions usually need a full reflash, because they change the kernel and Ubuntu release. Point releases within a line can often be applied as package updates on JetPack 5 and later. JetPack 7.2 also brings an officially supported Yocto path for Orin and Thor; our article on JetPack 7.2 and Yocto on Jetson Orin covers what that means for deployed fleets.

Many Jetson applications now ship as containers rather than packages installed straight onto the module. NVIDIA publishes Jetson base images on its NGC catalog, from a minimal Jetson Linux base to images with the full JetPack libraries or popular machine learning frameworks pre-installed.

The rule is that a container must match the Jetson Linux major release on the module, because the GPU driver lives on the host and is mounted into the container at run time. A container built for R35 will not run correctly on an R36 module. Tag images with the JetPack version they were built for, and upgrade the host and containers together.

Because JetPack is tied to a Jetson Linux version, the software and the module generation have to line up. A module that has reached the end of its software line, such as an original Jetson Nano on JetPack 4.6, keeps working but receives no new library versions. Our NVIDIA Jetson computers and rugged Jetson computers list the module in each system, which tells you which JetPack lines it can run. For a new design, choose a module with a long software runway as well as the performance you need.

We flash and configure the right JetPack release for the module, carrier board and application, so every system ships with a validated image as standard. Tell us your software stack, and we will match the JetPack version to it.