NVIDIA Jetson T4000 AI Computing Module with Blackwell GPU 1200 TFLOPS
The NVIDIA Jetson T4000 module delivers exceptional performance for physical AI and robotics applications at the edge. Powered by the advanced Blackwell GPU architecture, this system-on-module offers up to 1200 FP4 TFLOPS of AI compute and 64 GB of LPDDR5X memory with a configurable power envelope between 40 W and 70 W. Compared to the previous-generation AGX Orin, the Jetson T4000 delivers over 4x the AI compute performance with exceptional energy efficiency and real-time reasoning capabilities, making it ideal for humanoid robotics, industrial automation, autonomous machines, and edge AI deployments.
Unmatched AI Performance for Edge Applications
The Jetson T4000 features a 1536-core NVIDIA CUDA GPU based on the cutting-edge Blackwell architecture with 5th generation Tensor Cores, transformer engine support, and Multi-Instance GPU (MIG) capabilities. This allows it to effortlessly run the latest generative AI models, including Vision Language Action (VLA) models like NVIDIA Isaac GR00T for humanoids, as well as popular LLMs and VLMs such as Cosmos Reason. The module accelerates low-latency, real-time multi-sensor applications with a 12-core Arm Neoverse-V3AE CPU, delivering the computational power needed for sophisticated AI inference pipelines in manufacturing, logistics, retail, and healthcare.
Key Features
-
Blackwell GPU Architecture: 1536 CUDA cores with 5th gen Tensor Cores and Multi-Instance GPU support
-
1200 FP4 TFLOPS: Sparse AI compute performance optimized for generative AI workloads
-
64 GB LPDDR5X Memory: High-bandwidth memory with 273.2 GB/s bandwidth via 256-bit interface
-
12-Core Arm Neoverse-V3AE CPU: Powerful 64-bit CPU for real-time processing and control
-
3x 25 GbE Networking: High-speed connectivity for multi-sensor fusion and distributed AI
-
Configurable Power: 40 W to 70 W thermal design power for flexible deployment scenarios
-
Advanced Accelerators: Third-generation Programmable Vision Accelerator (PVA), dual video encoders/decoders, and optical flow accelerator
-
NVIDIA AI Software Stack: Runs NVIDIA Isaac for robotics, Metropolis for visual AI, and Holoscan for sensor processing
High-Speed Connectivity and I/O
The Jetson T4000 module is designed for sensor-rich applications requiring extensive I/O capabilities. It includes three 25 Gigabit Ethernet ports for high-speed networking, extensive USB support, and multiple interfaces for camera and sensor integration. The module supports PCIe Gen 5 for ultra-fast storage and peripheral connectivity, making it perfect for autonomous mobile robots (AMRs), smart infrastructure, industrial equipment, and machine vision systems.
Video Processing Capabilities
With 1x NVENC and 1x NVDEC hardware video codec engines, the Jetson T4000 enables real-time 4K and 8K video encoding and decoding across multiple streams. It supports H.265 (HEVC) and H.264 standards with impressive throughput including:
- Video Decode: 4x 8Kp30, 10x 4Kp60, 22x 4Kp30, 46x 1080p60, 92x 1080p30 (H.265)
- Video Encode: 6x 4Kp60, 12x 4Kp30, 24x 1080p60, 50x 1080p30 (H.265)
Designed for Production Deployment
The Jetson T4000's compact form factor (100 mm x 87 mm) and energy-efficient design make it ideal for integration into space-constrained and power-limited devices. The production-ready module is fully supported by NVIDIA JetPack 7.1 SDK, providing a comprehensive software ecosystem with AI frameworks, libraries, and development tools. This significantly shortens development and validation cycles for system integrators and equipment manufacturers bringing intelligent products to market.
Applications
- Humanoid robotics and autonomous mobile robots (AMRs)
- Industrial automation and smart factory systems
- Machine vision and visual inspection
- Edge AI inference and real-time analytics
- Smart city infrastructure and surveillance
- Medical imaging and healthcare devices
- Autonomous vehicles and drones
What's in the Box
- NVIDIA Jetson T4000 Module (SKU: 900-13834-0000-000)
- Product documentation
Note: The Jetson T4000 is a system-on-module and requires a compatible carrier board for operation. Developer kits and carrier boards are available separately.
Full Specifications
GPU
| Architecture |
NVIDIA Blackwell |
| CUDA Cores |
1536 |
| Tensor Cores |
5th Generation with Transformer Engine |
| AI Performance |
Up to 1200 FP4 TFLOPS (Sparse) |
| Multi-Instance GPU |
Yes |
| Graphics APIs |
OpenGL 4.6+, OpenGL ES 3.2, Vulkan 1.1+, CUDA 11.4+ |
CPU
| Architecture |
Arm Neoverse-V3AE (64-bit) |
| Cores |
12 |
Memory
| Type |
LPDDR5X |
| Capacity |
64 GB |
| Memory Interface |
256-bit |
| Bandwidth |
273.2 GB/s |
Video Processing
| Video Encoder (NVENC) |
1x instance, H.265 (HEVC) and H.264 support |
| Video Decoder (NVDEC) |
1x instance, H.265 (HEVC) and H.264 support |
| Decode Performance |
Up to 4x 8Kp30, 10x 4Kp60, 92x 1080p30 (H.265) |
| Encode Performance |
Up to 6x 4Kp60, 50x 1080p30 (H.265) |
Accelerators
| Programmable Vision Accelerator |
3rd Generation PVA |
| Optical Flow Accelerator |
Yes |
| JPEG Processing |
Hardware NVJPEG support |
Networking
| High-Speed Ethernet |
3x 25 Gigabit Ethernet (25 GbE) |
Power
| TDP Range |
40 W to 70 W (configurable) |
| Manufacturing Process |
3 nm (TSMC) |
Physical Specifications
| Form Factor |
System-on-Module (SOM) |
| Dimensions |
100 mm × 87 mm (module size) |
Software Support
| SDK |
NVIDIA JetPack 7.1+ |
| AI Frameworks |
NVIDIA Isaac, Metropolis, Holoscan, support for TensorFlow, PyTorch, ONNX Runtime |
| AI Models |
Isaac GR00T, Cosmos Reason, LLMs, VLMs, VLA models |
Specifications are sourced from the manufacturer product page and may be updated. Some specifications are preliminary and subject to change.
