NVIDIA Jetson Thor
Type: Platform Tags: NVIDIA, Jetson Thor, Jetson AGX Thor, Jetson T5000, Jetson T4000, Blackwell, physical AI, robotics, edge AI, JetPack Related: NVIDIA-Jetson-Platform, NVIDIA-JetPack-SDK, NVIDIA-Jetson-Linux, NVIDIA-Isaac-GR00T, NVIDIA-Isaac-ROS, NVIDIA-Blackwell-Architecture, NVIDIA-MIG, NVIDIA-DRIVE-AGX-Thor Sources: https://developer.nvidia.com/embedded/develop/hardware, https://developer.nvidia.com/embedded/buy-jetson, https://developer.nvidia.com/blog/?p=104879, https://docs.nvidia.com/jetson/archives/r38.4/DeveloperGuide/ Last Updated: 2026-04-29
Summary
NVIDIA Jetson Thor is the Blackwell-generation Jetson platform for physical AI, humanoid robotics, advanced medical systems, and autonomous machines. Public NVIDIA material describes Jetson AGX Thor / T5000-class hardware with up to 2070 FP4 TFLOPS, 128 GB memory, a 14-core Arm Neoverse-V3AE CPU, Blackwell Tensor Cores, MIG support, PVA v3, high-speed camera/network I/O, and a JetPack 7 software stack. This page consolidates Jetson Thor/T5000/T4000 information into one canonical topic rather than separate pages for each SKU.
Detail
Purpose
Jetson Thor brings data-center-generation AI capabilities into a power-constrained embedded module for real-time physical AI. It targets robots and autonomous machines that need local multimodal reasoning, perception, planning, control, and sensor processing without round-tripping to the cloud.
Key capabilities
- Blackwell-generation integrated GPU and Tensor Core stack for FP4/FP8 generative AI and VLA-style workloads.
- Up to 2070 FP4 TFLOPS and 128 GB memory on the top Thor/T5000-class configuration.
- Jetson T4000 tier with lower memory/performance/power targets for scaled physical AI deployments.
- MIG support for isolating workloads on a single embedded GPU, useful for mixed-criticality robot workloads.
- PVA v3, optical-flow acceleration, video encode/decode blocks, and high-speed camera connectivity for perception pipelines.
- QSFP, Ethernet, USB, PCIe, CAN, and other I/O on the developer kit for sensor fusion and robot prototyping.
- NVIDIA-JetPack-SDK 7.x and NVIDIA-Jetson-Linux 38.x software stack with Ubuntu 24.04/kernel 6.8-era platform direction.
NVIDIA context
Jetson Thor is the deployment endpoint for NVIDIA’s current physical AI stack. It connects NVIDIA-Isaac-GR00T humanoid foundation models, NVIDIA-Isaac-ROS deployment packages, NVIDIA-JetPack-SDK edge software, NVIDIA-Blackwell-Architecture compute capabilities, and NVIDIA-MIG workload partitioning in an embedded form factor.
Connections
- NVIDIA-Jetson-Platform - parent Jetson family page covering Thor, Orin, and earlier Jetson generations.
- NVIDIA-JetPack-SDK - software stack for Thor-class Jetson devices.
- NVIDIA-Jetson-Linux - Jetson Linux 38.x is the current Thor-focused BSP/software branch.
- NVIDIA-Isaac-GR00T - humanoid robotics workflows reference Jetson AGX Thor as robot-side compute.
- NVIDIA-Isaac-ROS - ROS 2 acceleration stack for Jetson robot deployment.
- NVIDIA-Blackwell-Architecture - Thor uses Blackwell-generation GPU technology.
- NVIDIA-MIG - Thor brings Multi-Instance GPU style isolation into the Jetson physical AI platform.
- NVIDIA-DRIVE-AGX-Thor - automotive Thor counterpart for DRIVE AGX development.
Source Excerpts
- NVIDIA developer pages position Jetson Thor modules as physical AI and robotics platforms with up to 2070 FP4 TFLOPS and 128 GB memory.
- NVIDIA’s Jetson Thor blog highlights Blackwell GPU architecture, MIG, PVA v3, and extensive I/O for humanoid robotics and sensor fusion.
- Jetson Linux 38.4 documentation lists Jetson Thor/T5000/T4000 support as part of the current Jetson Linux branch.