Overview
The NVIDIA Jetson T5000 is the top module of the Jetson Thor series, the platform NVIDIA built for physical AI and next-generation robotics. Its 2,560-core NVIDIA Blackwell GPU with 96 fifth-generation Tensor Cores delivers up to 2,070 TFLOPS of FP4 AI compute, roughly 7.5x the AI performance of Jetson AGX Orin, so generative AI, vision language action models and multi-sensor perception can run together on one module.
A 14-core Arm Neoverse-V3AE CPU, 128GB of LPDDR5X memory at 273 GB/s, four 25GbE ports and up to 20 camera inputs over the Holoscan Sensor Bridge make T5000 the compute heart of humanoid robots, surgical systems and autonomous machines. It is a production module: pair it with a carrier board of your own or from the NVIDIA partner ecosystem.
Key Features
- 2,070 FP4 TFLOPS: Blackwell GPU with 2,560 CUDA cores and 96 fifth-generation Tensor Cores, Multi-Instance GPU with 10 TPCs
- 14-core Neoverse-V3AE CPU: Arm v9 cores at up to 2.6 GHz with 1MB L2 per core and 16MB shared L3
- 128GB LPDDR5X: 256-bit interface, 273 GB/s of bandwidth for large multimodal models
- Rich sensor I/O: Up to 20 cameras over HSB, 16 MIPI CSI lanes, 4x 25GbE, PCIe Gen5, 3x USB 3.2
- Video engines: Encode up to 6x 4Kp60 H.265, decode up to 4x 8Kp30 H.265
- 40W to 130W: Configurable power envelopes with an integrated thermal transfer plate and heat pipe
- Full NVIDIA software stack: JetPack 7, Isaac, Holoscan, Metropolis and CUDA on a single module
Technical Specifications
| Specification | Details |
|---|---|
| AI Performance | 2,070 TFLOPS (FP4, sparse) |
| GPU | 2,560-core NVIDIA Blackwell GPU, 96 fifth-generation Tensor Cores, MIG with 10 TPCs |
| GPU Max Frequency | 1.57 GHz |
| CPU | 14-core Arm Neoverse-V3AE (64-bit), up to 2.6 GHz |
| Vision Accelerator | 1x PVA v3 |
| Memory | 128GB 256-bit LPDDR5X, 273 GB/s |
| Storage | NVMe over PCIe, SSD over USB 3.2 |
| Video Encode | 6x 4Kp60, 12x 4Kp30, 24x 1080p60, 50x 1080p30 (H.265) |
| Video Decode | 4x 8Kp30, 10x 4Kp60, 22x 4Kp30, 46x 1080p60 (H.265) |
| Camera | Up to 20 cameras over HSB, up to 6 cameras over 16 MIPI CSI lanes (C-PHY 2.1, D-PHY 2.1) |
| PCIe | Up to 8 lanes Gen5 (root port and endpoint) |
| USB | 3x USB 3.2, 4x USB 2.0 |
| Networking | 4x 25GbE |
| Display | 4x shared HDMI 2.1 / DisplayPort 1.4 (HBR2, MST) |
| Other I/O | 5x I2S, 2x AHUB, 2x DMIC, 4x UART, 4x CAN, 3x SPI, 13x I2C, 6x PWM |
| Power | 40W to 130W |
| Form Factor | 100 mm x 87 mm, 699-pin board-to-board connector, integrated TTP with heat pipe |
Ideal Use Cases
- Humanoid and mobile robots running vision language action models
- Surgical robotics and advanced medical devices
- Autonomous machines with many high-resolution sensors
- On-device generative AI and agentic workflows at the edge
- Industrial inspection with multi-camera perception
Why Choose This Product?
Jetson T5000 is the most capable embedded AI module NVIDIA has shipped: server-class Blackwell compute, 128GB of memory and the sensor bandwidth that modern robots need, in a single module with the complete NVIDIA robotics software stack. If your roadmap includes multimodal or generative AI at the edge, this is the platform to build on.
Interested? Contact us for pricing, volume orders and deployment support.
