Skip to main content

NVIDIA JetPack 7.0 for Jetson Thor White Paper

· 8 min read

1. Introduction​

NVIDIA JetPack 7.0 represents the most comprehensive and unified AI software platform for edge computing, purpose-built for the latest Jetson Thor series. It integrates a modern CUDA 13 environment, Ubuntu 24.04 LTS base, and Linux Kernel 6.8, delivering unmatched AI performance and real-time responsiveness across robotics, industrial, and medical domains.

As a long-term hardware partner of NVIDIA, Advantech enhances JetPack’s ecosystem through BSP optimization, validated camera modules, and industrial I/O integration. These enhancements simplify industrial deployment of NVIDIA Jetson systems while respecting NVIDIA’s software architecture and value.

2. JetPack 7.0 Overview​

JetPack 7.0 is the unified SDK for AI, robotics, and vision development across the NVIDIA Jetson platform family, built upon Jetson Linux 38.2, which incorporates Ubuntu 24.04 LTS and Linux Kernel 6.8. It connects seamlessly to NVIDIA’s ecosystem layers—Isaac, Metropolis, and Holoscan—enabling an end-to-end workflow from AI model training in DGX to simulation in Omniverse OVX and deployment on Jetson AGX.

Key highlights include:

  • Jetson Linux 38.2 (Ubuntu 24.04 LTS, Kernel 6.8) with real-time and PREEMPT_RT support.
  • CUDA 13.x, TensorRT 10.x, cuDNN 9.x, and Triton Inference Server 24.x.
  • DeepStream 7.0 SDK with multi-stream and multi-model inference.
  • Holoscan SDK v3.7.0 plus Holoscan Sensor Bridge for high-bandwidth sensor data fusion.
  • Containerization ready via NVIDIA Container Runtime (nvidia-docker) for GPU-accelerated application deployment and isolation.
  • AI stack support for LLM, VLM, VLA models, vision transformers, and CNN architectures.
  • Frameworks such as PyTorch, VLLM, and SGLANG, fully integrated with TensorRT and Triton.
  • Security framework spanning Secure Boot, TPM, TEE, Measured Boot, and encrypted storage.
  • ROS 2 Humble and Isaac ROS GEMs for robotics integration.
  • Support for Jetson AGX Thor and Jetson T5000 modules.

JetPack 7.0 unifies AI inference, real-time control, and cloud-native deployment, bridging the gap between training on NVIDIA DGX and executing on Jetson edge devices.

3. Jetson Linux and Industrial BSP Integration​

Jetson Linux 38.2 is the foundation of JetPack 7.0, providing kernel-level optimizations, device drivers, and security updates for all Jetson modules. It includes a pinmux configuration tool and table, allowing developers to flexibly map GPIO, I²C, SPI, CAN, and other interfaces to meet specific carrier board requirements.

Advantech has converted and optimized these pinmux settings for its industrial carrier boards, enabling out-of-box hardware functionality. If customers wish to redefine pin configurations, they can still use NVIDIA’s official pinmux tool for customized adaptation.

To streamline system deployment, Advantech provides a customized industrial BSP built upon Jetson Linux that:

  • Integrates industrial I/O, including RS-485, CAN, GPIO, watchdog, and power sequencing.
  • Includes wireless modules (Wi-Fi 6, 5G, BLE) validated with Jetson Linux 38.x.
  • Offers a unified Industrial API for monitoring, power, and watchdog control.
  • Remains fully compatible with NVIDIA’s JetPack SDK stack.

The BSP Launcher tool simplifies deployment and version management:

  • Flash JetPack-based images to Advantech hardware such as AIMB-294, AFE-A702, and AIR-075.
  • Build custom Jetson Linux images with optional driver packages.
  • Generate visualized SBOM and CVE reports for software transparency and compliance.
  • Provide quick JetPack version updates so users can maintain and upgrade their Jetson systems efficiently.

This integration allows customers to accelerate edge deployment while maintaining alignment with NVIDIA’s official ecosystem.

4. DeepStream, AI Inference, and Visualization Tools​

NVIDIA DeepStream 7.0 is the foundation for AI-powered video analytics on Jetson. It supports multiple neural networks, streaming sources, and sensor modalities in parallel with optimized data pipelines leveraging TensorRT, CUDA, and DLA cores.

With Triton Inference Server integration, JetPack 7.0 enables simultaneous deployment of heterogeneous models—from CNNs to transformers—while dynamically optimizing GPU resource allocation.

Advantech extends this ecosystem by offering a GUI-based AI Inference Demo Tool that:

  • Allows users to deploy DeepStream and TensorRT models without coding.
  • Provides visualization of inference results and frame latency.
  • Displays CPU, GPU, and DLA utilization in real time.

This bridges technical evaluation and business demonstration, helping customers rapidly assess JetPack’s AI capabilities in industrial environments.

5. Camera and Vision Solutions​

JetPack 7.0 delivers a robust vision stack supporting V4L2, Argus, and GStreamer, compatible with MIPI CSI and GMSL 2/3 interfaces. It also supports hardware-accelerated preprocessing through PVA (Programmable Vision Accelerator) and IPS pipelines for real-time image transformation.

Advantech collaborates with industrial camera vendors to offer pre-validated GMSL camera solutions, ensuring consistent performance and synchronized operation with Jetson Thor. Through the integration, users can control sensor synchronization, triggering, and exposure alignment, complementing NVIDIA’s Holoscan Sensor Bridge and Isaac ROS for high-precision data capture and time alignment.

6. Robotics, ROS 2, and Isaac Integration​

JetPack 7.0 is fully integrated with ROS 2 Humble, Isaac ROS GEMs, and NVIDIA Isaac Sim or Isaac Lab, enabling simulation-to-reality workflows. Developers can leverage GPU-accelerated SLAM, 3D perception, path planning, and motion control through Isaac and DeepStream.

Advantech extends this with ROS nodes for industrial interfaces such as digital I/O, UART, CAN, GPIO, and PWM, simplifying integration between robots and industrial automation systems. Through interoperability with NVIDIA Omniverse OVX and DGX AI platforms, JetPack users can perform model training with TAO or NIM, optimization through CUOpt, and real-time simulation, then deploy results directly to Jetson hardware for a consistent AI pipeline from cloud to edge.

7. Time Synchronization and Industrial Networking​

JetPack 7.0 provides native support for IEEE 1588 PTP and NVIDIA Time Sync, offering sub-microsecond precision across multiple sensors and compute nodes. When combined with the real-time kernel (PREEMPT_RT), this ensures deterministic scheduling for industrial control, multi-camera robotics, and autonomous systems.

Advantech carrier boards implement hardware timestamping, ensuring accurate synchronization between GPU, CPU, and peripheral components, which enables reliable edge AI in time-critical industrial networks.

8. Cloud-Native Deployment and Containerization​

JetPack 7.0 leverages container technology for modern edge AI deployment. Built upon the unified CUDA 13 ecosystem, the platform includes Docker and the NVIDIA Container Runtime, which expose GPU acceleration, DLA, and multimedia hardware directly to the container environment so models trained on DGX can be deployed seamlessly to Jetson.

Key advantages for industrial deployment, validated by Advantech, include:

  • Portability and isolation: NVIDIA NGC containers ensure environment consistency (build once, deploy anywhere), eliminating dependency conflicts across heterogeneous systems.
  • Simplified MLOps: Robust CI/CD pipelines enable rapid updates, rollbacks, and scalable management of AI applications at the edge.

Advantech enhances this cloud-native capability by providing NVIDIA AI Enterprise (NVAIE) services. As an official partner, Advantech ensures:

  • Certified platforms: Edge AI systems are NVAIE-certified, delivering the optimized performance and stability required for mission-critical deployments.
  • Enterprise support: Tier-1 support and consulting services for NVAIE software, including NVIDIA NIM microservices, accelerate time-to-market for Generative AI and Vision AI deployments.

9. Security and Compliance​

Security and reliability are foundational pillars of JetPack 7.0. NVIDIA Jetson Linux 38.2 includes Secure Boot, Measured Boot, Trusted Execution Environment (TEE), memory and disk encryption, firmware TPM, secure storage, and verified SBOM with CVE tracking.

Advantech aligns with these capabilities through industrial certification frameworks such as IEC 62443 VOC (Verification of Conformity) and EN 18031 DOC (Declaration of Conformity). This ensures customers deploying JetPack-based industrial systems meet global cybersecurity and operational reliability standards.

10. Ubuntu Licensing and Deployment Support​

JetPack 7.0’s foundation on Ubuntu 24.04 LTS introduces commercial licensing considerations for large-scale OEM deployment. Advantech provides a validated Ubuntu licensing solution that allows customers to legally preload and clone Ubuntu-based JetPack images on production systems, ensuring compliance with Canonical’s commercial terms.

This service simplifies mass production and factory provisioning for Jetson-based devices, ensuring smooth delivery and consistent image integrity.

11. Conclusion​

NVIDIA JetPack 7.0 defines the next generation of edge AI by unifying GPU acceleration, real-time performance, and robotics intelligence into a single ecosystem. It connects seamlessly with NVIDIA DGX for AI training and Omniverse OVX for simulation, and extends through Holoscan, Isaac, and Metropolis for domain-specific edge AI deployment.

Through Advantech’s industrial enhancements—including BSP Launcher (which enables quick JetPack version updates), Industrial API, camera validation, and compliance documentation—customers can deploy JetPack systems faster, safer, and with full industrial compatibility. Together, NVIDIA and Advantech are enabling a robust, synchronized, and secure edge AI infrastructure that bridges industrial reliability with cutting-edge GPU intelligence.

image_1763082933923.png

© 2025 NVIDIA Corporation & Advantech Co., Ltd. All rights reserved.