A comprehensive development platform for the NVIDIA Jetson series, JetPack 6.0 forms the central basis for AI-supported applications in edge computing. The Software Development Kit (SDK) brings together vital components such as CUDA, TensorRT, cuDNN and DeepStream to create a powerful, all-in-one package – perfect for embedded AI, robotics and industrial image processing.
The transition from JetPack 5.x to 6.0 represents a significant technological advancement. With a new system base, improved hardware utilisation and modern container integration, JetPack 6.0 enhances performance and developer convenience, opening up new possibilities for efficient, scalable AI projects at the edge.
The all-in-one toolkit for AI development on Jetson
JetPack is a comprehensive solution for AI development on NVIDIA Jetson systems. Included are a Linux distribution (previously based on Ubuntu 18.04/20.04, now 22.04), CUDA for GPU computing, cuDNN for deep learning, TensorRT for optimising models for inference, OpenCV for image processing and DeepStream for video analytics. It provides a powerful foundation for AI-powered edge computing projects for both beginners and professionals.
More performance, flexibility and future-proofing
Version 6.0 brings a significant change to the technology stack:
- New operating system: JetPack 6.0 is based on Ubuntu 22.04 LTS for the first time, enabling modern kernel features and better software compatibility.
- Kernel upgrade: Linux Kernel 5.15 brings security and performance improvements.
- CUDA 12.2: Enables new features for parallel processing and improved memory management.
- TensorRT 8.6 provides accelerated inference support for the latest deep learning models, including transformer-based networks.
- Improved container support: JetPack 6.0 is fully containerised, greatly simplifying development, testing, and deployment.
Additionally, JetPack 6.0 enables the simultaneous use of multiple Jetson boards with different JetPack versions for the first time, making it extremely helpful for large-scale projects involving legacy components.
JetPack 6.0 shows its capabilities on the new Jetson generation
Not all Jetson boards benefit from JetPack 6.0: Older boards, such as the Jetson Nano or TX2, are no longer supported. JetPack 6.0 is therefore primarily designed for the current Orin generation. This allows the full potential of the hardware to be realised.Performance gain: Initial benchmarks show a performance improvement of up to 30% for typical AI inference tasks. Boot times, energy efficiency, and camera handling have also improved. TensorRT 8.6 also delivers significantly better results with transformer models such as BERT and YOLOv7.
The following are currently compatible:
- Jetson AGX Orin
- Jetson Orin NX
- Jetson Orin Nano
- Jetson Xavier NX (limited compatibility, available only via special BSPs)
ReComputer series from Seeed: JetPack 6.0 makes edge AI even more powerful
The ReComputer series from Seeed Studio is based on Jetson boards (mostly Orin or Xavier) and is of particular interest to developers thanks to its plug-and-play systems with extensive I/O.Many ReComputer models with Orin Nano or Orin NX are JetPack 6.0 compatible. The upgrade offers several advantages:
- Better container integration for edge AI
- Newer drivers for sensors and peripherals
- Optimised CUDA support for demanding inference applications
Seeed Studio already offers JetPack 6.0-compatible images and manual flashing assistance for some models.
How to upgrade
There are two ways to carry out the upgrade process:
- Via the NVIDIA SDK Manager: This is a convenient GUI for flashing the Jetson device via USB. This will reset the device.
- Manual flashing using tools such as flash.sh: This gives more control to advanced users and is useful for headless installations.
Please note: The upgrade will reset the system, meaning that all data will be lost if no backup is available. In many cases, a new installation is recommended over an in-place upgrade, as JetPack 6.0 introduces significant changes.
Greater productivity for developers through optimised AI workflow
JetPack 6.0 brings not only system and performance improvements, but also real productivity gains.
- Better Devcontainer compatibility: Development in isolated, portable environments.
- Updated Python and AI frameworks: Full compatibility with the latest versions of PyTorch, TensorFlow and ONNX.
- DeepStream 6.4 support: Complex video AI pipelines are enabled out of the box.
- Remote debugging and profiling tools: Improved development user experience for professional workflows.
Real-time AI for transportation, industry, and medicine
JetPack 6.0’s potential is particularly evident in applications with high computing requirements:
- Autonomous vehicles: Better processing of multi-sensor inputs with lower latency.
- Smart cities: Real-time video analysis on Jetson Orin with significantly higher FPS rates.
- Industrial robotics: More responsive control through optimised CUDA utilisation.
- Medical image processing: Improved diagnostic algorithms on compact edge devices.
JetPack 6.0 paves the way for the next generation of Jetson
development. Users of Orin-based systems will benefit significantly from the new software. In addition to improved performance and developer convenience, stability and future-proofing are also key features.
However, for users with older boards, the move also means saying goodbye to the past. JetPack 6.0 is a clear commitment to the Orin generation. Anyone planning future-proof projects should switch now.
Outlook: NVIDIA will continue to develop the Jetson platform towards modular containerisation, remote development, and automated AI pipelines. JetPack 6.0 marks the beginning of a new era in development.
Images: NVIDIA, Seeed, Adobe Stock
Update – JetPack 6.1:
NVIDIA has delivered a significant performance and stability boost for all Jetson Orin-based devices with JetPack 6.1.
Particularly noteworthy:
- Up to 40% less CPU load for camera applications – thanks to optimised video processing and GStreamer improvements.
- New fTPM security features – for a trusted platform with better protection at the hardware level.
- MAXN mode for Jetson Orin Nano – for the first time, the maximum power level (15 W) is also available for compact projects.
- Updated CUDA and TensorRT stack – for even stronger AI inference performance and more efficient use of computing units.
- Uniform network interface names – for more stable system configurations and better script compatibility.
