Deep dive into Jetpack 6 Developer Preview: bring your own kernel, upgradable compute stack, and many more!

NVIDIA recently released the much-anticipated Jetpack 6, an exciting development marking further advancements in edge AI solution development. Essentially, JetPack can be considered a combination of Jetson Linux (L4T, BSP + drivers, and some basic packages) and JetPack components (CUDA, cuDNN, TensorRT packages, etc.). You can selectively install these JetPack components or even uninstall Ubuntu GUI-related packages to achieve a minimal configuration. 

 Let’s dive deeper into it now! 

Jetpack Roadmap

Jetpack 4 is set to release its final version, 4.6.1, in 2024 and will then reach the end of its lifecycle. Customers using Jetson Xavier NX are highly recommended to upgrade to Jetpack 5 for ongoing support. Starting with the next version, Jetpack 5 will focus on security fixes and critical updates. Meanwhile, future Jetpack 6 versions will introduce new features and computational updates, with a lifecycle mainly spanning five years.

Jetpack 6 Release 

Highlights

• Jetpack 6 brings greater flexibility in Linux kernel selection (Kernel 5.15 and up) and Jetson Linux 36.2 which packs Linux Kernel 5.15 and Ubuntu 22.04 based root file system.
• Developers now have the freedom to use custom kernels with versions 5.0 or higher.
• The release also includes a wider choice of Linux-based distributions from partners like Canonical and the latest CUDA-X compute stack, which can be upgraded independently of Jetpack.

The developer preview of JetPack 6 introduces kernel 5.15, CUDA 12.2, and TensorRT 8.6. However, it’s not yet ready for production and lacks security features and OTA capabilities. These will be included in the major production version, set for release in March 2024. JetPack 6.1 is scheduled for Q3 2024, and in the interim, JetPack 5 versions will continue to be released.

Bring Your Own Kernel

NVIDIA has been contributing to the upstream kernel for several years to support commercial Linux distributions. To use your own kernel, you’ll need to identify the initial list of patches and build a set of out-of-tree modules for drivers. It’s also crucial to review the kernel in the official NVIDIA release for additional necessary patches.

Linux Distros on Jetson

JetPack 6 includes Jetson Linux out of the box, but Jetson ecosystem partners offer various Linux-based Distros on Jetson, including:

• Canonical Ubuntu: Ubuntu 22.04 is offered by Canonical.
• Red Hat: Red Hat Enterprise Linux 9.3 offered by Red Hat
• Windriver: Commercial embedded Linux OS from Windriver.
• RedHawk: Real Time Linux OS from Concurrent Real-Time for mission-critical hard real-time applications.
• Yocto-based distros and services leveraging meta-Tegra are offered by ecosystem partners like
•  Balena, Foundaries.io & TimeSys.

Upgradable Compute Stack

JetPack 6 introduces an upgradable compute stack, a significant advancement for developing Jetson-powered systems. Traditionally, system teams prepare for production by validating Board Support Packages (BSPs) and enabling security and OTA features. Simultaneously, application teams work on AI models and applications using development kits, often requiring the latest AI stack, like the newest TensorRT versions.

Previously, accessing the latest AI stack meant migrating to the newest JetPack version, tying the compute stack to the BSP and creating a dependency. JetPack 6 resolves this by separating the compute stack from the BSP. This change allows application teams to upgrade to the latest AI compute stack independently, without waiting for BSP updates, thereby streamlining development and reducing dependencies.

Migrate Jetpack 6 from JetPack 5

When migrating, note the kernel change from 5.10 to 5.15 and the filesystem update. The Trusted OS remains unchanged, ensuring consistency in power performance and clocking. Camera drivers are now packaged as “out-of-tree” drivers, with minimal changes in multimedia, graphics, and security.

For production cases, to manage on-site Jetson device fleets, NVIDIA will provide image-based OTA tools, with which you can switch between different major versions. Customers we also have a range of device management tools offered by our ecosystem partners, which can be integrated with these tools to provide the same functionality.  

Based on our tests, here are some issues you might encounter when flashing the image. The Jetpack developer preview is not yet a production version, and developers’ suggestions for maintaining the BSP for Seeed Jetson devices are crucial. We look forward to your feedback.

System Usability and Flashing 

There are some usability and flashing issues in the Jetpack 6 Developer Preview. For instance, Jetson AGX Orin development kits might not resume correctly after being paused, and Jetson devices using the rtl8822ce WiFi driver might experience TLP packet format errors. To resolve this, please restart the device.  

For flashing issues, updating the qemu-user-static package on the Linux host used for installation is recommended.

Camera 

When using IMX219 or IMX477 cameras with Jetson Orin NX and Jetson Orin Nano devices, it’s essential to specify the correct sensor-id in the gstreamer nvarguscamera element.

For better stability and performance, note that JetPack 6.0 is a developer preview release and not intended for production. It is ready to start your development on Jetson Orin with the JetPack 6 software stack. This release does not include any security features or OTA features. The production quality release of JetPack 6 is targeted for March 2024 and will include security and OTA features along with a production-quality stack. Please carefully read the JetPack 6.0 Developer Preview Release Notes for known issues and other details. Seeed Studio’s technical support team is available to provide professional guidance 

Build Gen AI Applications Based on Jetpack 6

By combining the Transformer model and large language model, generative AI goes beyond the traditional model based on convolutional neural network and can detect, segment, track, and search more flexibly. Based on the understanding of human language, it significantly improves the ease of use to improve the model and simplify the development, deployment, and management of edge AI. Generative models and use cases can now be experienced on the Generative AI Lab, which also supports Jetpack 6 and provides tutorials on text generation, LLaVa, vision transformers, image/video search, and more.

Metropolis Microservice: a unique suite of cloud-native microservices for vision AI solution building.

Metropolis Microservice seamlessly integrates AI capabilities, including generative AI and multimodal approaches, into Jetson applications to deliver vision AI-driven solutions for enterprise users, to concur complex tasks such as multi-camera tracking and few-shot learning, to improve critical operational efficiency and safety issues.

This platform brings a powerful set of APIs and microservices for developers to quickly develop complex vision-based applications. With the Metropolis on Jetson end-to-end workflow, LLM (Large Language Model) microservices can run in the cloud or on Jetson for easy interaction with natural language prompts. Users can run advanced zero-sample inference models and update them dynamically using prompts. 

Beyond the pre-defined app structure and microservice configuration, developers have the flexibility to modify pre-trained models, APIs, and even the microservice code itself. No-code tools like NVIDIA TAO further streamline the process, making it easy to retrain AI models. It’s time to leverage these potent and customizable cloud-native building blocks to craft your next Vision AI solution.

Seeed NVIDIA Jetson Ecosystem

Seeed is an Elite partner for edge AI in the NVIDIA Partner Network. Explore more carrier boards, full system devices, customization services, use cases, and developer tools on Seeed’s NVIDIA Jetson ecosystem page.

Join the forefront of AI innovation with us! Harness the power of cutting-edge hardware and technology to revolutionize the deployment of machine learning in the real world across industries. Be a part of our mission to provide developers and enterprises with the best ML solutions available. Check out our success case study catalog to discover more edge AI possibilities!

Take the first step and send us an email at [email protected] to become a part of this exciting journey! 

Download our latest Jetson product Catalog to find one option that suits you well. If you can’t find the off-the-shelf Jetson hardware solution for your needs, please check out our customization services, and submit a new product inquiry to us at [email protected] for evaluation

About Author