Skip to main content

JetPack 7.2 Resource Hub

JetPack 7.2 Resource Hub cover

This page organizes the JetPack 7.2 resources currently available for Seeed Studio NVIDIA Jetson products. Use it as the starting point for understanding the new software stack, selecting a supported image, migrating an existing JetPack 6.x project, restoring peripheral drivers, and deploying optimized AI workloads.

info

JetPack 7.2 uses Jetson Linux 39.2, an Ubuntu 24.04 root filesystem, and Linux kernel 6.8. NVIDIA Jetson Linux 39.2 supports both the Jetson Orin and Jetson Thor platform families.

What Is New in JetPack 7.2?

JetPack 7.2 is more than an operating-system update. It extends the JetPack 7 software architecture to Jetson Orin and adds platform capabilities for agentic AI, production Linux customization, memory optimization, and higher-performance edge inference.

Agentic AI and Developer Workflows

  • NVIDIA NemoClaw readiness: JetPack 7.2 provides the required platform dependencies for one-command NemoClaw installation and local or cloud model orchestration.
  • Jetson agent skills: NVIDIA provides reusable device-side and BSP-side workflows for Jetson Linux customization, memory optimization, model benchmarking, package selection, and application prototyping.
  • Cloud-native deployment: The JetPack stack continues to support containerized development and deployment workflows for edge services.

Platform Architecture

  • Jetson Orin support in JetPack 7: JetPack 7.2 brings the JetPack 7 software stack to the Jetson Orin family while retaining Jetson Thor support.
  • Unified ISO installation: Jetson Linux 39.2 introduces a unified ISO-based installation path for supported Jetson Orin and Jetson Thor developer kits.
  • SBSA alignment: Jetson Thor follows the Server Base System Architecture software model, improving portability across Arm server-class platforms.

Yocto

  • Official Yocto Project support: NVIDIA-validated OpenEmbedded/Yocto recipes provide a path to reproducible, customized, and production-oriented Linux images.

Performance, Memory, and Isolation

  • Jetson AGX Orin 32GB Super Mode: JetPack 7.2 adds the MAXN_SUPER power mode for supported Jetson AGX Orin 32GB configurations.
  • Memory-efficiency workflows: Jetson agent skills can audit and reduce bootloader carveouts, kernel reservations, and unnecessary user-space memory consumption.
  • Multi-Instance GPU on Jetson Thor: MIG is available as a technology preview on supported Jetson Thor T5000 configurations for isolated multi-workload execution.
warning

Some JetPack 7.2 features are platform-specific. MIG and SBSA-specific behavior apply to Jetson Thor, while MAXN_SUPER applies to supported Jetson AGX Orin 32GB configurations. Confirm the module, carrier board, BSP, power supply, and thermal design before enabling a new power or acceleration mode.

JetPack 7.2 Software Baseline

LayerJetPack 7.2 baselineMigration impact
Jetson Linux39.2Rebuild out-of-tree kernel modules and BSP customizations.
Root filesystemUbuntu 24.04Revalidate package names, repositories, Python environments, and system services.
Linux kernel6.8Rebuild camera, Wi-Fi, fieldbus, and custom peripheral drivers against the new headers.
CUDA generationCUDA 13Rebuild CUDA applications and do not reuse JetPack 6.x binaries without validation.
TensorRT enginesJetPack 7.2 TensorRT stackRebuild serialized TensorRT engines on the target software stack.
Supported platformsJetson Orin and Jetson ThorUse the correct BSP, toolchain flags, and precision support for the target GPU architecture.

JetPack 7.2 Ecosystem Map

This collection is intentionally limited to JetPack 7.2 material. Existing articles are copied into the JetPack_7_2 tree and use an _bk slug so the series can be reviewed, updated, translated, and released independently from the original Wiki pages.

JetPack 7.2 capabilityIncluded resource
Unified Orin and Thor software architectureUnified Platform, ISO, and SBSA (planned)
Ubuntu 24.04, Linux 6.8, and CUDA 13 migrationJetPack 7.2 Deep Dive and Migration Playbook (planned)
Agentic AI and reusable Jetson skillsRapid Prototyping with NVIDIA Skills and NemoClaw on Jetson Thor
YoctoBuild and Flash a Yocto Image
Higher-performance inference with TensorRT Edge-LLMDeploy TensorRT Edge-LLM on JetPack 7.2
Lower system and LLM memory useJetPack 7.2 Memory Optimization
DeepStream 9.1 and natural-language video workflowsDeepStream on JetPack 7.2
AGX Orin MAXN_SUPER and Thor MIGMAXN_SUPER and MIG (planned)
Kernel 6.8 driver transitionJetPack 7.2 Wireless Module Setup and Camera and Multimedia Compatibility (planned)

Flash & OTA

ResourceCoverage
JetPack 7.2 Deep DivePlatform changes, JetPack 6.2 comparison, migration impact, and AGX Orin inference results.
Flash and OTA Upgrade to JetPack 7.2Clean flashing, image-based OTA requirements, version verification, and deployment choice.
JetPack 6.x to JetPack 7.2 Migration PlaybookPlanned: backup, dependency rebuild, acceptance testing, rollback, and fleet migration.
JetPack 7.2 Unified Platform, ISO, and SBSAPlanned: unified installation, Orin/Thor differences, and Thor SBSA behavior.
JetPack 7.2 MAXN_SUPER and MIGPlanned: AGX Orin performance mode and Thor workload isolation.

Kernel 6.8 Drivers and Multimedia

ResourceCoverage
JetPack 7.2 Wireless Module Setup GuideJetPack 7.2 driver and firmware recovery for Intel AX210/AX200 and Realtek RTL8852BE.
JetPack 7.2 Camera and Multimedia CompatibilityPlanned: CSI, GMSL, Argus, V4L2, GStreamer, codecs, and multi-camera validation.
warning

JetPack 6.x kernel modules, camera drivers, device-tree binaries, and TensorRT engines must not be reused directly on JetPack 7.2. Rebuild them against the Jetson Linux 39.2 software stack.

Agentic AI and Jetson Skills

JetPack 7.2 expands the Jetson developer workflow beyond manual setup by making reusable agent skills and local agentic applications first-class parts of the ecosystem.

ResourceEcosystem role
Rapid Prototyping on Jetson with NVIDIA SkillsDevice inspection, compatibility checks, memory analysis, environment preparation, prototype construction, and packaging.
Control reBot Arm B601 with NemoClaw on Jetson ThorLocal perception, LLM reasoning, tool execution, service management, and physical-AI control on the JetPack 7 platform.

Memory Efficiency

ResourceCoverage
JetPack 7.2 Memory OptimizationSkills-based auditing, headless/no-camera BSP reclamation, SWIOTLB safety, quantization, KV-cache control, and lower-memory LLM inference.

Yocto

ResourceEcosystem role
Build and Flash a Yocto Image for reComputer SuperReproducible OpenEmbedded/Yocto image construction for a production-oriented Jetson Linux deployment.

AI Deployment & Applications

ResourceCoverage
Deploy TensorRT Edge-LLM on JetPack 7.2JetPack 7.2 model export, Orin/Thor build targets, engine generation, and C++ inference.
Industrial Vision Monitoring on JetPack 7.2YOLO and VLM monitoring verified on reComputer Industrial and reServer Industrial with L4T 39.2.
DeepStream on JetPack 7.2DeepStream 9.1 installation, agentic skills, natural-language pipeline authoring, VLM integration, migration, and memory planning.
Deploy Full-Weight GR00T N1.7 on JetPack 7.2 and AGX OrinValidated seven-engine TensorRT deployment, numerical verification, offline inference, and portable path configuration for AGX Orin.

Serialized engines and custom TensorRT plugins must be rebuilt on the target JetPack 7.2 software stack.

Coverage Still Reserved

JetPack 7.2 featureReserved page
Full JetPack 6.x migration and rollbackMigration Playbook
Unified ISO, Orin/Thor split, and SBSAUnified Platform, ISO, and SBSA
MAXN_SUPER, MIG, and performance isolationMAXN_SUPER and MIG
CSI/GMSL and accelerated multimediaCamera and Multimedia Compatibility
  1. Confirm that the target Seeed product has a JetPack 7.2 BSP or image.
  2. Back up application data, calibration files, container volumes, and custom device-tree sources.
  3. Flash JetPack 7.2 and validate boot, storage, networking, and recovery mode.
  4. Restore Wi-Fi, camera, CAN, EtherCAT, or other out-of-tree drivers with JetPack 7.2 builds.
  5. Rebuild CUDA applications, TensorRT plugins, and TensorRT engines.
  6. Validate the application in the original power mode before enabling MAXN_SUPER or other performance modes.
  7. Record memory use, thermals, power consumption, latency, and throughput before moving the device into production.

Official NVIDIA References

Tech Support & Product Discussion

Thank you for choosing our products! We are here to provide you with different support to ensure that your experience with our products is as smooth as possible. We offer several communication channels to cater to different preferences and needs.

Loading Comments...