NVIDIA Jetson AGX Thor Development Kit 128GB
NVIDIA Jetson AGX Thor Development Kit
945-14070-0085-000
- NVIDIA Jetson T5000 module with heat sink and reference carrier board
- 140W DC power
- 802.11ax wireless network interface controller
- 1 TB NVMe installed in the M.2 Key-M slot
- Quick Start Guide
The Definitive Platform for Humanoid Robotics
The NVIDIA Jetson AGX Thor Developer Kit represents the most advanced platform for humanoid robotics, offering unparalleled performance and scalability. Equipped with the NVIDIA Blackwell GPU and a robust 128 GB of memory, the platform delivers up to 2070 FP4 TFLOPS of AI computing performance. This enables it to run the latest generative AI models effortlessly within a modern, powerful chassis. Compared with the NVIDIA Jetson AGX Orin, the Jetson AGX Thor provides up to 7.5x more AI computing capability and 3.5x greater energy efficiency.
The Jetson Thor is designed to accelerate low-latency, real-time applications through innovative Blackwell Multi-Instance GPU (MIG) technology and a powerful 14-core Arm Neoverse-V3AE CPU. It also includes a comprehensive set of accelerators, such as the third-generation Programmable Vision Accelerator (PVA), dual encoders and decoders, an optical flow accelerator, and more. For high-speed sensor fusion, the development kit offers a wide range of I/O options, including a QSFP port with 4x25GbE, a wired RJ45 Multi-GbE connector, multiple USB ports, and additional connectivity interfaces. Its design enables seamless integration with existing humanoid robot platforms, providing straightforward connectivity to accelerate prototyping.
The Jetson Thor introduces a new class of robotic computers architected from the ground up to power the next generation of humanoid robots. It supports a wide range of generative AI models, from Vision-Language-Action (VLA) models such as NVIDIA Isaac GR00T N to popular LLMs (Large Language Models) and VLMs (Vision Language Models). To provide a seamless cloud-to-edge experience, Jetson Thor runs the NVIDIA AI software stack for physical AI applications, including NVIDIA Isaac for robotics, NVIDIA Metropolis for visual agent AI, and NVIDIA Holoscan for sensor processing. It also enables the development of AI agents at the Edge using NVIDIA agentic AI workflows such as Video Search and Summarization (VSS).
Our partner ecosystem provides all the carrier boards, design services, cameras, and other sensors you need, as well as additional AI and system software to accelerate solution development across industries such as robotics, smart spaces, retail, manufacturing, medicine, and more.
Its massive AI computing capability and sensor-processing features in a compact form factor make the Jetson AGX Thor Developer Kit the ideal platform for developers seeking to unlock new possibilities in humanoid robotics and other physical AI applications.
|
|
Jetson AGX Thor Developer Kit |
Jetson T5000 Module |
|
AI Performance |
2070 TFLOPS (FP4—Sparse) |
||
GPU |
2560-core NVIDIA Blackwell architecture GPU with 96 fifth-generation Tensor Cores Multi-Instance GPU (MIG) with 10 TPCs |
||
GPU Max Frequency |
1.57 GHz |
||
CPU |
14-core Arm® Neoverse®-V3AE 64-bit CPU 1 MB L2 cache per core 16 MB shared system L3 cache |
||
CPU Max Frequency |
2.6 GHz |
||
Vision Accelerator |
1 PVA v3 |
||
Memory |
128 GB 256-bit LPDDR5X 273 GB/s |
||
Storage |
1 TB NVMe M.2 Key M slot |
Supports NVMe through PCIe Supports SSD through USB3.2 |
|
Video Encode |
6x 4Kp60 (H.265) 12x 4Kp30 (H.265) 24x 1080p60 (H.265) 50x 1080p30 (H.265) 48x 1080p30 (H.264) 6x 4Kp60 (H.264) |
||
Video Decode |
4x 8Kp30 (H.265) 10x 4Kp60 (H.265) 22x 4Kp30 (H.265) 46x 1080p60 (H.265) 92x 1080p30 (H.265) 82x 1080p30 (H.264) 4x 4Kp60 (H.264) |
||
Camera |
HSB camera via QSFP slot USB camera |
Up to 20 cameras via HSB Up to 6 cameras through 16x MIPI CSI-2 lanes Up to 32 cameras using Virtual Channels C-PHY 2.1 (10.25 Gbps) D-PHY 2.1 (40 Gbps) |
|
PCIe* |
M.2 Key M slot with x4 PCIe Gen5 M.2 Key E slot with x1 PCIe Gen5 |
Up to Gen5 (x8 lanes) Root port only—C1 (x1) and C3 (x2) Root port or endpoint—C2 (x1), C4 (x8), and C5 (x4) |
|
USB* |
2x USB-A (3.2 Gen2) 2x USB-C (3.1) |
xHCI host controller with integrated PHY (up to) 3x USB 3.2 4x USB 2.0 |
|
Networking* |
1x 5GBe RJ45 connector 1x QSFP28 (4x 25 GbE) |
4x 25 GbE |
|
Display |
1x HDMI 2.0b 1x DisplayPort 1.4a |
4x shared HDMI2.1 VESA DisplayPort 1.4a—HBR2, MST |
|
Other I/O |
QSFP connector M.2 Key E expansion slot (WLAN/BT, x1 PCIe, USB2.0, UART, I2C, I2S) M.2 Key M connector (NVMe for storage) PCIe x4 lane, I2C, PCIe x2 lane 2x 13-pin CAN headers 2x 6-pin automation headers LED JTAG connector (2x 5-pin header) 1x fan connector—12V, PWM, and Tach Audio panel header (2x 5-pin) Microfit power connector 2-pin RTC backup battery connector |
5x I2S/2x audio hub (AHUB), 2x DMIS, 4x UART, 4x CAN, 3x SPI, 13x I2C, 6x PWM outputs |
|
Power |
40W – 130W |
||
Mechanical |
243.19 mm x 112.40 mm x 56.88 mm Thermal Transfer Plate (TTP) and optional fan or heat sink |
100 mm x 87 mm 699-pin B2B connector Integrated Thermal Transfer Plate (TTP) with heatpipe |
|
* Refer to the Software Features section of the latest NVIDIA Jetson Linux Developer Guide for a list of supported features.
** Low-speed I/O specification is subject to change.
Extreme AI Performance for Robotics
Powered by the Jetson T5000 module, the Blackwell GPU with 2560 cores and 96 Tensor Cores delivers up to 2070 TFLOPS FP4 / 1035 TFLOPS FP8 — ideal for running a wide range of generative AI models such as LLM, VLA (including NVIDIA® Issac™ GR00T N), ViT, etc.
Optimised for Transformer and Multimodal Workloads
The integrated Transformer engine, MIG (Multi-Instance GPU) support, and 128 GB of LPDDR5X memory (273 GB/s) enable efficient training and inference at the Edge.
Real-Time Processing and Control
It features a 14-core Arm® Neoverse®-V3AE CPU for fast, deterministic performance in real-time robotic applications.
High-Speed Sensor and Network Integration
Supports up to 4 25 GbE links, a 140 W power supply, and a comprehensive set of I/O interfaces, including HDMI, DP, USB 3.2/USB-C, CAN, and GbE, for high-bandwidth, low-latency real-time sensor fusion.
