NVIDIA Jetson AGX Thor T5000 AI Computing Module for Physical AI and Robotics
The NVIDIA Jetson T5000 module is the most powerful edge AI computing platform ever created, delivering 2070 FP4 TFLOPS of AI compute and 128 GB of LPDDR5X memory in a compact system-on-module (SoM) form factor. Powered by the groundbreaking Blackwell GPU architecture with 2560 CUDA cores and 96 fifth-generation Tensor Cores, this module is designed for the next generation of physical AI applications, including humanoid robotics, autonomous machines, advanced medical systems, and intelligent edge devices. With configurable power between 40W and 130W, the T5000 delivers 7.5 times the AI performance and 3.5 times better energy efficiency than the previous-generation Jetson AGX Orin, making it the ultimate platform for real-time generative AI at the edge.
Why Choose the NVIDIA Jetson T5000?
-
Unmatched AI Performance: 2070 FP4 TFLOPS powered by NVIDIA Blackwell architecture GPU with Multi-Instance GPU (MIG) support and Transformer Engine for running the latest generative AI models including Vision Language Action (VLA) models, LLMs, and VLMs
-
Massive Memory Bandwidth: 128 GB of 256-bit LPDDR5X memory with 273 GB/s bandwidth for handling complex multimodal sensor processing and large AI models
-
High-Performance CPU: 14-core Arm Neoverse V3AE 64-bit CPU at up to 2.6 GHz, delivering 2.6 times the performance of previous-generation cores
-
Advanced Networking: 4x 25 GbE and 1x 5 GbE connectivity for high-speed sensor fusion and real-time data processing in robotics applications
-
Extensive I/O Options: PCIe Gen5 support (12 lanes at 32 GT/s per lane), third-generation Programmable Vision Accelerator (PVA v3.0), dual video encoders/decoders, and optical flow accelerator
-
Production-Ready: Compact 100mm x 87mm form factor compatible with existing AGX carrier boards, 5-year operating lifetime (24x7), and support from NVIDIA's comprehensive AI software stack
Blackwell GPU Architecture for Physical AI
The NVIDIA Blackwell GPU at the heart of the Jetson T5000 represents a generational leap in edge AI computing. With 2560 CUDA cores and 96 fifth-generation Tensor Cores, this GPU delivers up to 7.8 FP32 TFLOPS and introduces native FP4 quantization with a next-generation Transformer Engine. The Multi-Instance GPU (MIG) technology enables predictable performance in mixed-criticality robotics applications, allowing you to partition GPU resources for simultaneous workloads. This makes the T5000 ideal for running sophisticated AI models like NVIDIA Isaac GR00T for humanoid robots, large language models for natural language interaction, and vision language models for real-time scene understanding.
14-Core Arm Neoverse CPU for Real-Time Processing
The Jetson T5000 features a powerful 14-core Arm Neoverse V3AE CPU running at up to 2.6 GHz, providing 2.6 times the performance of the 12-core Orin CPU. Each core includes 64KB + 64KB L1 cache, 1MB L2 cache, and shares a 16MB L3 cache for efficient data access. This high-performance CPU architecture is optimised for real-time control, sensor processing, and running complex robotics software stacks alongside the GPU workloads.
Multi-Sensor Fusion and High-Speed Networking
Designed for advanced robotics and autonomous systems, the T5000 excels at processing data from multiple sensors simultaneously. The module supports 4x 25 GbE networking plus 1x 5 GbE, enabling high-bandwidth connectivity for lidar, radar, cameras, and other sensors. The third-generation Programmable Vision Accelerator (PVA) handles image signal processing for up to 24-megapixel sensors, while dual video encoders and decoders support H.265 (HEVC), H.264, VP9, AV1, and more. An integrated optical flow accelerator and Holoscan Sensor Bridge further enhance real-time multi-sensor processing capabilities.
Comprehensive AI Software Stack
The Jetson T5000 runs JetPack 7 with Ubuntu 24.04 LTS and CUDA 13.0, providing a unified development experience from cloud to edge. It supports the complete NVIDIA AI software stack for physical AI applications, including:
-
NVIDIA Isaac for robotics development and deployment
-
NVIDIA Metropolis for visual agentic AI and intelligent video analytics
-
NVIDIA Holoscan for high-performance sensor processing
- Support for popular AI frameworks and models through Jetson AI Lab
- Agentic AI workflows like Video Search and Summarisation (VSS) for edge deployment
Applications Across Industries
The NVIDIA Jetson T5000 module is trusted by leading organisations across multiple sectors:
-
Humanoid Robotics: Powers next-generation humanoid robots from companies like Figure, Agility Robotics, and Boston Dynamics
-
Manufacturing & Logistics: Enables intelligent automation and warehouse robotics for Amazon Robotics and others
-
Autonomous Machines: Drives autonomous vehicles and heavy equipment for Caterpillar and agricultural applications
-
Medical Devices: Supports advanced medical imaging and surgical robotics systems from Medtronic
-
Smart Spaces: Powers intelligent building systems, retail analytics, and visual AI applications
-
Research & Development: Used by Stanford, Carnegie Mellon, and other leading universities for robotics research
Compact Form Factor with Production-Grade Reliability
Despite its enormous computing power, the Jetson T5000 maintains a compact 100mm x 87mm form factor, identical to the existing AGX modules. This ensures compatibility with a wide ecosystem of carrier boards and enclosures from partners like Connect Tech, Aetina, and others. The module is designed for 5-year operating lifetime (24x7) with an operating temperature range of -25°C to 115°C (Tj) and thermal throttling protection. The module weighs just 0.28 kg without packaging, making it ideal for space-constrained robotics and edge AI deployments.
Configurable Power for Every Application
The Jetson T5000 offers flexible power configurations from 40W to 130W, with a default power mode of 120W. Custom power modes can be created using the Jetson Power Estimator to optimise performance and efficiency for your specific application. The module operates from a 7V-20V input voltage range, making it suitable for battery-powered autonomous systems and industrial deployments.
What's in the Box
- NVIDIA Jetson T5000 Module (P/N: 900-13834-0080-000)
- Quick start documentation
Full Specifications
GPU
| Architecture |
NVIDIA Blackwell |
| CUDA Cores |
2560 |
| Tensor Cores |
96 (5th generation) |
| AI Performance |
2070 FP4 TFLOPS (Sparse), 1035 INT8 TOPS |
| FP32 Performance |
7.8 TFLOPS |
| Multi-Instance GPU |
Yes (MIG support) |
CPU
| Architecture |
14-core Arm Neoverse V3AE (ARMv9 64-bit) |
| Max Frequency |
2.6 GHz |
| L1 Cache per Core |
64 KB instruction + 64 KB data |
| L2 Cache per Core |
1 MB |
| L3 Cache (shared) |
16 MB |
Memory
| Capacity |
128 GB LPDDR5X |
| Bus Width |
256-bit |
| Bandwidth |
273 GB/s |
Video Processing
| Video Encoders |
2x (H.265/HEVC, H.264) |
| Video Decoders |
2x (H.265, H.264, VP9, VP8, AV1, MPEG-4, MPEG-2, VC-1) |
| Image Signal Processors |
2x ISP (up to 24 MP sensors) |
| Programmable Vision Accelerator |
PVA v3.0 |
| Optical Flow Accelerator |
Yes |
Connectivity
| Ethernet |
4x 25 GbE + 1x 5 GbE |
| PCIe |
12 lanes Gen5 (32 GT/s per lane) |
Power
| Power Range |
40W to 130W (configurable) |
| Default Power Mode |
120W |
| Input Voltage |
7V-20V (SYS_VIN_HV) |
Physical
| Form Factor |
100mm x 87mm (AGX compatible) |
| Weight (module only) |
0.28 kg |
Environmental
| Operating Temperature (Tj) |
-25°C to 115°C |
| Thermal Transfer Plate (TTP) Max |
80°C |
| Throttle Temperature |
109°C |
| Operating Lifetime |
5 years (24x7 operation) |
Software
| Operating System |
Ubuntu 24.04 LTS |
| SDK |
JetPack 7 |
| CUDA Version |
13.0 |
| AI Software |
NVIDIA Isaac, Metropolis, Holoscan, Jetson AI Lab |
Specifications are sourced from the manufacturer product page and may be updated.
