NVIDIA Jetson Orin - JetPack 7.2 Components
🚧 Documentation is under development
The NVIDIA Jetson Orin guide is currently under active development. Some sections may be incomplete or change without notice.
Questions? Contact RidgeRun or email to support@ridgerun.com.
- Introduction
- JetPack 7.2
- JetPack 6.1
- JetPack 5.0.2
-
Development in the Board
-
Getting into the Board
-
- Emulation
- Supported Cameras
- GStreamer Pipelines
- RidgeRun Products
- RidgeRun Reference Designs
- Reference Documentation
- Contact Us
This section describes the main software components of JetPack 7.2 for NVIDIA Jetson AGX Orin. JetPack 7.2 supports the Jetson Orin family and uses Jetson Linux 39.2.
JetPack Components
| Component | Jetpack 6.1 Version | Jetpack 7.2 Version |
|---|---|---|
| Jetson Linux (L4T) | 36.4 | 39.2 |
| Linux kernel | 5.15 | 6.8 |
| Ubuntu-based root file system | 22.04 | 24.04 |
| CUDA Toolkit | 12.6 | 13.2.1 |
| cuDNN | 9.3 | 9.20.0 |
| TensorRT | 10.3 | 10.16.2 |
| Vision Programming Interface (VPI) | 3.2 | 4.1.3 |
| Programmable Vision Accelerator (PVA) software | Not specified in NVIDIA's release summary | 2.9.1 |
Linux for Tegra
Jetson Linux supplies the operating system, NVIDIA drivers, firmware, and bootloader. Release 39.2.0 includes a Linux 6.8 kernel, a UEFI bootloader, and an Ubuntu 24.04-based root file system. Its supported target architecture is aarch64; NVIDIA lists Ubuntu 22.04 and 24.04 as Linux host distributions for flashing, and GCC 13.2 as the cross-compilation toolchain.[1]
For Orin users, this release adds the Orin family to JetPack 7 and introduces MAXN_SUPER support for Jetson AGX Orin 32GB. It also provides a unified ISO installation method for Orin and Thor developer kits and official OpenEmbedded/Yocto recipes. Multi-Instance GPU (MIG) is a technology preview for Thor T5000; it is not supported on the Jetson Orin series in this release.[1]
Multimedia and Camera APIs
The Jetson camera stack includes libargus for low-level camera control, nvarguscamerasrc for GStreamer capture with ISP control through Argus, and V4L2 for direct capture. The Multimedia API includes samples for previewing, capturing, and recording camera streams.[2]
JetPack 7.2 also includes Jetson SIPL API Package 2.0.0. NVIDIA documents GMSL and Camera-over-Ethernet (CoE) support, unified camera/query samples, and stereo pipelines. Argus remains supported, while NVIDIA directs new camera framework development toward SIPL. Camera and ISP capabilities depend on the platform and framework; consult NVIDIA's camera support documentation for the intended sensor and board.[1][3]
CUDA
JetPack 7.2 includes CUDA Toolkit 13.2.1.[4] CUDA provides a C/C++ compiler, runtime, GPU libraries, and development tools for GPU-accelerated applications.[5]
CUDA libraries and tools have independent version numbers. Table 2 lists selected versions from NVIDIA's CUDA Toolkit 13.2 Update 1 release notes; it is a toolkit reference, rather than a list of every package installed on a Jetson.[6]
| Component | Version |
|---|---|
| CUDA Runtime (cudart) | 13.2.75 |
| CUDA NVCC | 13.2.78 |
| CUDA NVRTC | 13.2.78 |
| CUPTI | 13.2.75 |
| Compute Sanitizer API | 13.2.76 |
| cuBLAS | 13.4.0.1 |
| cuFFT | 12.2.0.46 |
| cuRAND | 10.4.2.55 |
| cuSOLVER | 12.2.0.1 |
| cuSPARSE | 12.7.10.1 |
| NPP | 13.1.0.48 |
| nvJPEG | 13.1.0.48 |
| Thrust, CUB, and libcu++ | 3.2.0 |
CUDA Deep Neural Network Library (cuDNN)
JetPack 7.2 includes cuDNN 9.20.0.[4] cuDNN supplies GPU-accelerated deep learning operations, including convolution, attention, matrix multiplication, pooling, and normalization. NVIDIA provides frontend and backend APIs for building operation graphs and executing these computations.[7]
TensorRT
JetPack 7.2 includes TensorRT 10.16.2.[4] TensorRT provides engine compilation and runtime APIs for optimized inference. Models can be imported through ONNX, and NVIDIA provides C++ and Python interfaces. Use the support matrix for the selected TensorRT release when checking framework and dependency compatibility.[8][9]
OpenCV
NVIDIA documents OpenCV interoperability in VPI, allowing computer vision applications to combine OpenCV with NVIDIA acceleration backends.[10] NVIDIA's JetPack 7.2 release summary does not list a specific OpenCV version, so this page does not assign an OpenCV version to the release.[4]
Vision Programming Interface (VPI)
JetPack 7.2 includes VPI 4.1.3.[4] VPI offers computer vision and image processing algorithms through C/C++ and Python interfaces, with CPU, CUDA, PVA, VIC, and OFA backends depending on the algorithm and hardware.[10]
VPI 4.1.3 adds Orin support in the VPI 4.1 branch and compatibility with CUDA 13.2. Its updates include custom CUDA work submission in VPI stream order, NvSciBuf image interoperability, and additional PVA implementations such as Canny edge detection, bilateral filtering, and perspective warp.[11]
References
- ↑ 1.0 1.1 1.2 NVIDIA Jetson Linux 39.2.0 GA Release Notes
- ↑ NVIDIA Jetson Linux 39.2: Camera Software Development Solution
- ↑ NVIDIA Jetson Linux 39.2: Camera Development
- ↑ 4.0 4.1 4.2 4.3 4.4 NVIDIA JetPack 7.2 downloads and release information
- ↑ NVIDIA CUDA Toolkit 13.2.1 documentation
- ↑ 6.0 6.1 NVIDIA CUDA Toolkit 13.2 Update 1 Release Notes
- ↑ NVIDIA cuDNN
- ↑ NVIDIA TensorRT: Architecture Overview
- ↑ NVIDIA TensorRT 10.x Support Matrix
- ↑ 10.0 10.1 NVIDIA VPI 4.1 documentation
- ↑ NVIDIA VPI 4.1.3 Release Notes