NVIDIA Jetson T4000 Module - High-Performance AI Computing for Physical AI and Robotics
The NVIDIA Jetson T4000 Module delivers exceptional performance for physical AI and robotics applications, powered by the advanced Blackwell GPU architecture. With up to 1200 FP4 TFLOPS of AI compute and 64GB of LPDDR5x memory, this compact system-on-module (SoM) provides 4× the AI performance of the previous generation Jetson AGX Orin, all within a 40W to 70W configurable power envelope. The T4000 is optimised for tighter power and thermal requirements while delivering the real-time reasoning and multi-sensor processing capabilities essential for next-generation autonomous machines, humanoid robotics, and edge AI deployments.
Why Choose the NVIDIA Jetson T4000?
-
4× AI Performance Improvement: Over 1200 FP4 TFLOPS of compute power compared to Jetson AGX Orin for advanced AI workloads
-
Energy-Efficient Design: Configurable power profiles from 40W to 70W with 3.5× better energy efficiency than previous generation
-
Blackwell GPU Architecture: Features Transformer engine and Multi-Instance GPU (MIG) support for generative AI models
-
64GB High-Speed Memory: LPDDR5x memory provides the capacity needed for large language models (LLMs) and vision language models (VLMs)
-
12-Core Arm Neoverse V3AE CPU: Low-latency processing for real-time multi-sensor applications and sensor fusion
-
Production-Ready Ecosystem: Supported by JetPack 7.1 software, extensive partner carrier boards, and NVIDIA Isaac robotics platform
Advanced AI Capabilities
The Jetson T4000 is designed for physical AI applications that require real-time reasoning, multimodal perception, and sophisticated decision-making. It supports a wide range of generative AI models, including Vision Language Action (VLA) models like NVIDIA Isaac GR00T for humanoid robots, as well as popular large language models and vision language models such as Cosmos Reason. The module features hardware acceleration for AI inference with TensorRT Edge-LLM SDK, enabling efficient deployment of LLMs and VLMs directly at the edge.
High-Performance Video Processing
Equipped with 1× NVENC and 1× NVDEC hardware video codec engines, the T4000 enables real-time 4K video encoding and decoding. It supports H.265 (HEVC) and H.264 video standards, making it ideal for applications requiring advanced vision processing, video analytics, and multi-camera sensor fusion. The module also includes a third-generation Programmable Vision Accelerator (PVA) and optical flow accelerator for demanding computer vision workloads.
Comprehensive Connectivity and I/O
The Jetson T4000 accelerates low-latency, real-time multi-sensor applications with extensive I/O capabilities:
-
3× 25 GbE Networking: High-speed network connectivity for industrial and robotics applications
-
PCIe 5.0 Support: High-bandwidth interface for peripherals and expansion
-
MIPI CSI Camera Support: Advanced camera pipeline for multi-camera vision systems
-
GMSL Support: Automotive-grade camera connectivity for demanding real-time vision workloads
-
Extensive Low-Speed I/O: Multiple GPIO, I2C, SPI, UART interfaces for sensor integration
Ideal Applications
The NVIDIA Jetson T4000 is the perfect choice for developers and system integrators building:
-
Humanoid Robotics: Real-time control and perception for next-generation robots
-
Autonomous Mobile Robots (AMRs): Navigation, obstacle avoidance, and decision-making at the edge
-
Industrial Automation: Smart factories, quality inspection, and predictive maintenance
-
Smart Infrastructure: Video analytics, visual AI agents, and intelligent monitoring systems
-
Medical Robotics: Advanced surgical systems and medical imaging applications
-
Retail & Logistics: Inventory management, automated warehousing, and intelligent checkout systems
Software Ecosystem
The T4000 is supported by NVIDIA JetPack 7.1, the most advanced software platform for Jetson, featuring Linux Kernel 6.8 and Ubuntu 24.04 LTS. JetPack 7 aligns with Server Base System Architecture (SBSA) and includes support for CUDA 13.0, TensorRT, cuDNN, and comprehensive AI libraries. Developers gain access to:
-
NVIDIA Isaac Platform: Complete robotics development framework including Isaac GR00T foundational models
-
NVIDIA Metropolis: Visual agentic AI and intelligent video analytics
-
NVIDIA Holoscan: High-performance sensor processing and streaming
-
Jetson AI Lab: Open-source models, tutorials, and community resources
-
Partner Ecosystem: Production-ready carrier boards, cameras, sensors, and design services from established partners
What's in the Box
- NVIDIA Jetson T4000 Module
- Thermal interface material (TIM)
- Module mounting hardware
- Quick start documentation
Full Specifications
GPU
| Architecture |
NVIDIA Blackwell |
| GPU Cores |
2 GPC, 6 TPC, 1536 NVIDIA CUDA cores |
| Tensor Cores |
5th Generation Tensor cores with Transformer engine |
| AI Performance |
Up to 1200 FP4 TFLOPS (sparse) |
| Multi-Instance GPU |
MIG support for workload isolation |
CPU
| CPU Cores |
12-core Arm Neoverse V3AE (64-bit) |
| Cache |
50MB L2 cache |
Memory
| Memory |
64GB LPDDR5x |
| Memory Bandwidth |
Up to 273.2GB/s |
Video Processing
| Video Encode (NVENC) |
1× encoder (H.265, H.264) |
| Video Decode (NVDEC) |
1× decoder (H.265, H.264, 8K/4K support) |
| Vision Accelerator |
3rd generation Programmable Vision Accelerator (PVA) |
| Optical Flow Accelerator |
Hardware-accelerated optical flow and stereo disparity |
Networking & Storage
| Ethernet |
3× 25 GbE |
| PCIe |
PCIe 5.0 x16 |
Camera & Sensor Interfaces
| MIPI CSI |
Multiple camera lanes (specifications preliminary) |
| GMSL |
Automotive camera support |
Power
| Power Consumption |
Configurable 40W to 70W |
Software Support
| JetPack Version |
JetPack 7.1 |
| Operating System |
Ubuntu 24.04 LTS |
| Linux Kernel |
6.8 |
| CUDA |
CUDA 13.0+ |
| Graphics APIs |
OpenGL 4.6+, OpenGL ES 3.2, Vulkan 1.1+ |
Form Factor
| Module Type |
System-on-Module (SoM) |
| Interface |
Jetson Thor Series connector |
Specifications are sourced from the manufacturer product page and may be updated. Some specifications for the Jetson T4000 module encode, decode, and low-speed I/O are preliminary and subject to change.
