As AI agents transfer from the digital world to the bodily setting, they’ll readily use NVIDIA Jetson to speed up real-world deployment with optimized reminiscence and efficiency.
NVIDIA JetPack 7.2 instantly helps one-command deployment of NVIDIA NemoClaw, an open supply stack that provides privateness and safety controls to OpenClaw. It introduces NVIDIA agent expertise for Jetson—Jetson device-side skills and Jetson BSP skills—and extends the newest compute stack and agentic capabilities to NVIDIA Jetson Orin. The Jetson software-defined platform makes this attainable: the identical {hardware} continues to ship extra worth with each software program launch.
This put up introduces new JetPack 7.2 launch options and capabilities, which additionally embrace:
- NVIDIA Multi-Instance GPU (MIG) help on NVIDIA Jetson Thor for deterministic multiworkload execution
- Official Yocto Undertaking help for customized Linux distributions that may additional enhance system effectivity
- Tremendous Mode for Jetson AGX Orin 32 GB for increased AI efficiency and higher value effectivity on the edge
Collectively, these updates assist builders get extra out of current Jetson {hardware}, speed up time to market, and decrease complete value of possession.
How is NVIDIA JetPack 7.2 software program agentic-ready?
With JetPack 7.2, Jetson is NemoClaw-ready out of the field. JetPack 7.2 comes preconfigured with the required dependencies and software program stack, so you possibly can deploy and run NemoClaw-based workflows on Jetson with out guide setting setup. This allows you to simply construct agentic physical AI purposes throughout robotics, industrial automation, imaginative and prescient brokers, and edge AI techniques.
To put in NemoClaw on a Jetson system operating JetPack 7.2, run the next single command:
curl -fsSL nvidia.com/nemoclaw.sh | bash
NVIDIA agent expertise for Jetson in JetPack 7.2
JetPack 7.2 additionally gives builders with Jetson agent expertise to construct and optimize Jetson software program stacks utilizing AI brokers. The agent expertise are a set of repeatable, agent-executable directions that outline which instruments to name, what outputs to provide, and how you can validate outcomes. Quite than manually configuring every step of the event course of, builders can leverage agent expertise by an agent to deal with these duties mechanically.
Jetson agent expertise apply this sample particularly to Jetson software program growth workflows. These agent-driven workflows assist automate frequent growth duties corresponding to Jetson Linux customization, reminiscence optimization, mannequin benchmarking, and deployment configuration. With each device-side and BSP-side implementations, builders can use agent expertise to scale back growth complexity and speed up the trail from prototyping to manufacturing deployment on Jetson platforms.
JetPack 7.2 ships three classes of expertise:
- Jetson Linux customization expertise: Information an agent to construct and customise a BSP from scratch for customized provider boards. This consists of configuring I/Os, clock settings, fan management, energy profiles, or another module for a particular {hardware} design. Duties that beforehand required weeks of guide effort will be dealt with by an agent, decreasing time to marketplace for customized Jetson designs.
- Reminiscence optimization expertise: Optimize reminiscence utilization throughout the software program stack. These expertise can tune the entire stack beginning bootloader reminiscence carveouts, optimize kernel reminiscence reservation, cut back redundant person house processes, and assist construct essentially the most memory-efficient software program configuration for a given workload. This instantly reduces TCO by enabling extra succesful workloads to run on decrease reminiscence configurations.
- Mannequin benchmarking expertise: Make it easier to determine the perfect mannequin configuration in your use case. These expertise cowl mannequin benchmarking, inference optimization, and Jetson diagnostics. For instance, a developer constructing a NemoClaw-based utility can use these expertise to find out which mannequin runs most effectively on their goal system for his or her particular process.
Together with these three classes of expertise, NVIDIA can also be introducing expertise that assist brokers construct imaginative and prescient pipelines utilizing NVIDIA DeepStream and NVIDIA Metropolis Blueprint for Video Search and Summarization (VSS).
To study extra and get began, take a look at Jetson device-side skills and Jetson BSP skills on GitHub.
MIG on Jetson Thor allows GPU partitioning for mixed-criticality workloads
JetPack 7.2 on Jetson Thor introduces help for MIG, permitting the built-in NVIDIA Blackwell GPU to be partitioned into two remoted GPU situations with devoted compute, cache, and reminiscence bandwidth. This permits a number of AI workloads to run concurrently with predictable efficiency and minimal interference.
Mixed with the Preemptible RT kernel in JetPack 7, MIG helps create a extra deterministic execution setting for mixed-criticality techniques. Workload determinism is crucial for bodily AI techniques corresponding to humanoid robots, autonomous machines, industrial automation, and medical devices. It’s because notion, planning, management, generative AI, and safety workloads usually share a single SoC, the place useful resource rivalry can introduce latency jitter into time-sensitive pipelines.
With MIG on Jetson Thor, builders can dedicate GPU sources to latency-sensitive robotics workloads whereas operating best-effort AI inference or generative AI fashions on a separate partition. This helps keep predictable latency and high quality of service for workloads corresponding to notion, sensor fusion, movement planning, and security monitoring. JetPack 7.2 helps two MIG partitions on Jetson Thor:
- A bigger AI and graphics partition for inferencing, rendering, visualization, and common NVIDIA CUDA workloads (12 SMs, 1536 CUDA cores)
- A second remoted compute partition for robotics, management, notion, or safety-critical workloads (8 SMs, 1024 CUDA cores)
Purposes, containers, and companies will be assigned to particular MIG partitions utilizing commonplace CUDA Runtime controls and NVIDIA Container Toolkit integration. That is particularly necessary for next-generation humanoid robotics operating a number of AI pipelines throughout completely different timing domains, the place management loops, AI notion, and generative AI reasoning should reliably coexist on a single embedded platform.
By bringing data-center-class GPU partitioning to embedded AI computing, JetPack 7.2 allows extra succesful edge AI techniques with improved predictability and reliability for real-world deployment. Read more about MIG on Jetson Thor.
Introducing Yocto Undertaking help on NVIDIA Jetson
Beginning with JetPack 7.2, NVIDIA gives official Yocto Project support on Jetson, together with validated recipes and reference pictures for Jetson developer kits. The Yocto Project is an open supply Linux Basis venture that gives instruments to construct customized Linux distributions for embedded {hardware} architectures.
NVIDIA now leads roadmap contributions with a daily launch cadence to the OE4T layer. NVIDIA owns the CI/CD pipeline, SQA, and releases validated reference pictures for Jetson developer kits. And builders have entry to technical documentation and devoted boards help.
The Yocto Undertaking brings three core advantages to Jetson builders:
- Customizability: Lets you construct tightly tailor-made pictures that embrace solely the required companies, drivers, and libraries, reasonably than adapting the NVIDIA Ubuntu L4T picture. This reduces reminiscence footprint and optimizes system efficiency for the goal utility.
- Reproducibility: Yocto Undertaking produces an identical picture builds throughout runs, simplifying debugging, testing, and certification workflows. That is particularly precious in regulated fields corresponding to medical and industrial deployments.
- Open ecosystem. Entry hundreds of recipes and group layers for AI frameworks, industrial protocols, and customized middleware.
That can assist you resolve when to make use of L4T/JetPack versus OE4T/Yocto Undertaking, seek advice from the Developer Choice Information in Determine 3.
With the official help of Yocto Undertaking on Jetson, NVIDIA has additionally constructed a sturdy ecosystem of distribution companions, ISVs, and ODMs to speed up and simplify Yocto Undertaking growth on Jetson platforms. These companions present a variety of choices together with production-ready Linux distributions, BSP customization, long-term help, fleet administration options, multimedia and ISP experience, and security-focused integrations.
Corporations corresponding to Konsulko Group and Peridio provide full OS options like Konsulko Orca OS and Avocado OS, whereas Balena focuses on container-based fleet administration and deployment at scale. Different NVIDIA companions embrace Neurealm, RidgeRun, and Wind River, who present in depth engineering and NRE companies with deep experience in embedded Linux, BSP customization, multimedia pipelines, and long-term platform help. Collectively, this ecosystem allows builders to quickly deploy, customise, and scale Yocto-based options on Jetson.
Along with distribution companions and ISVs, NVIDIA additionally works intently with a powerful ecosystem of companions to assist clients speed up product growth and deployment on Jetson platforms corresponding to AAEON, Advantech, Antmicro, ASUS, AVerMedia, Connect Tech, EDOM, and YUAN present a variety of {hardware} options together with provider boards, edge AI techniques, industrial embedded platforms, video seize options, and reference designs optimized for Jetson. These companions allow builders to quickly prototype and scale production-ready AI and edge computing options with {hardware} platforms tailor-made for robotics, industrial automation, sensible cities, healthcare, retail, and different embedded AI purposes.
Unifying the Jetson stack and unlocking extra efficiency
JetPack 7.2 extends the Ubuntu 24.04, kernel 6.8 and CUDA Toolkit 13.0-based compute stack (launched with Jetson Thor) to the Jetson Orin household, bringing each platforms onto a single unified software program basis. With a typical stack throughout Orin and Thor, you possibly can seamlessly deploy the newest AI purposes throughout your entire Jetson portfolio whereas profiting from the latest CUDA capabilities, libraries, and efficiency optimizations.
This unified method considerably reduces the engineering effort required to help a number of {hardware} platforms, simplifying utility growth, validation, deployment, and long-term fleet upkeep.
JetPack 7.2 additionally introduces a brand new Tremendous Mode for Jetson AGX Orin 32 GB, unlocking increased GPU and energy configurations that carry its efficiency a lot nearer to Jetson AGX Orin 64 GB. By rising GPU frequencies from 930 MHz to 1.3 GHz and enabling increased energy envelopes as much as 60W, Tremendous Mode boosts AI efficiency from 200 TOPS to 241 TOPS, a greater than 20% improve over the usual AGX Orin 32 GB configuration.
This enhancement allows clients to realize near-flagship AGX Orin 64 GB efficiency utilizing Jetson AGX Orin 32 GB, whereas decreasing module value by 45%. The brand new Tremendous Mode makes the 32 GB module a cheap selection for generative AI, robotics, and edge AI deployments.
| Jetson AGX Orin 32 GB | Jetson AGX Orin 32GB Tremendous | Jetson AGX Orin 64 GB | |
| Nemotron3 Nano 30B A3B | 31 | 37 | 40 |
| Cosmos Motive 2 8B | 9 | 10 | 10 |
| Qwen 3.5 4B | 24 | 27 | 28 |
| Qwen 3.5 9B | 13 | 15 | 17 |
| Qwen 3.6 27B | 4 | 5 | 7 |
| Gemma 4 E4B | 25 | 29 | 32 |
Get began with NVIDIA JetPack 7.2
JetPack 7.2 delivers extra worth from the identical Jetson {hardware} by software program. As agentic AI strikes to the sting and reminiscence prices stay an actual constraint in manufacturing deployments, this launch instantly addresses each.
Options embrace one-command deployment of NVIDIA NemoClaw, reminiscence and workflow optimization agent expertise for Jetson, official Yocto Undertaking help for lean and reproducible manufacturing builds, and MIG on Jetson Thor for deterministic multiworkload execution. With JetPack 7.2, you are able to do extra on current {hardware} whereas constructing towards more and more succesful agentic workloads on the edge.
Obtain JetPack 7.2 to get began deploying agentic AI on the edge. For questions and group help, go to the NVIDIA Developer Discussion board.
Be part of NVIDIA founder and CEO Jensen Huang for the NVIDIA GTC Taipei 2026 Keynote and study extra with associated sessions.