NVIDIA IGX T5000 Blackwell AI Module 2070 TFLOPS Edge Computing
The NVIDIA IGX T5000 is a high-performance System-on-Module (SoM) engineered for the most demanding industrial, medical, and mission-critical edge AI applications. Powered by the latest NVIDIA Blackwell GPU architecture with 2,070 FP4 TFLOPS of AI compute, a 14-core Arm Neoverse-V3AE CPU, and 128 GB LPDDR5X memory, the IGX T5000 delivers exceptional performance for generative AI, robotics, and real-time sensor processing. This production-ready module combines enterprise-grade hardware with functional safety compliance (ISO 26262, IEC 61508, targeting ASIL D/SIL 3), making it ideal for autonomous machines, collaborative robotics, humanoid robots, and advanced medical imaging systems.
Unmatched AI Performance for Physical AI Applications
The IGX T5000 module features 2,560 CUDA cores with 96 fifth-generation Tensor Cores supporting FP4, FP8, FP16, bfloat16, TF32, and INT8 formats. Compared to NVIDIA AGX Orin, it delivers over 7.5× higher AI compute with 3.5× better energy efficiency. The integrated Transformer Engine and Multi-Instance GPU (MIG) support enable effortless deployment of multiple generative AI models, including Large Language Models (LLMs), Vision-Language Models (VLMs), and Vision-Language-Action models (VLAs) directly at the edge.
Key Features
-
Blackwell GPU Architecture: 2,560 CUDA cores with 96 fifth-generation Tensor Cores delivering up to 2,070 FP4 TFLOPS
-
14-Core Arm Neoverse-V3AE CPU: High-performance 64-bit processor for deterministic real-time control
-
128 GB LPDDR5X Memory: 256-bit memory interface with 273 GB/s bandwidth for data-intensive workloads
-
Functional Safety Island (FSI): 4 Cortex-R52 safety cores in Dual-Core Lock-Step (DCLS) mode with Hardware Security Module (HSM)
-
Advanced Sensor Processing: Supports up to 32 camera streams via 16-lane MIPI CSI-2 interface (164 Gbps C-PHY v2.1 / 40 Gbps D-PHY v2.1)
-
Video Processing: Dual NVDEC/NVENC engines for hardware-accelerated 8K30 and multiple 4K60 video streams
-
Programmable Vision Accelerator (PVA v3.0): Dedicated vector processor for computer vision and sensor data processing
-
Flexible Power Configuration: Configurable 40–130 W power envelope
-
10-Year Lifecycle Support: Enterprise-grade longevity with NVIDIA AI Enterprise software stack
-
Compact Embedded Form Factor: Designed for integration into custom carrier boards
Industrial-Grade Hardware for Mission-Critical Environments
The IGX T5000 is purpose-built for industrial and medical environments where reliability and safety are paramount. The integrated Functional Safety Island enables ISO 26262 and IEC 61508 compliance, targeting ASIL D and SIL 3 certification. The module's industrial-grade hardware design ensures reliable operation in harsh conditions, making it suitable for factory automation, autonomous mobile robots (AMRs), surgical robotics, autonomous vehicles, and warehouse automation.
Comprehensive Connectivity and I/O
The IGX T5000 offers extensive I/O options for demanding sensor-fusion applications:
-
High-Speed Networking: Support for 4× 25 GbE networking
-
Multi-Camera Support: 16-lane MIPI CSI-2 interface enabling up to 32 simultaneous camera data streams
-
Holoscan Sensor Bridge (HSB): Connectivity for up to 20 cameras
-
PCIe Gen5: Expanded high-speed I/O for custom carrier board designs
-
Flexible GPIO and Interfaces: Customizable I/O through carrier board integration
Enterprise AI Software Stack
The IGX T5000 runs the complete NVIDIA AI Enterprise software stack, providing a seamless cloud-to-edge experience. Supported frameworks include:
-
NVIDIA Isaac for robotics and autonomous machines
-
NVIDIA Metropolis for visual AI agents and video analytics
-
NVIDIA Holoscan for real-time sensor processing in medical and industrial applications
-
NVIDIA Halos Outside-in Safety Agent Blueprint for industrial safety AI agents
-
NVIDIA NIM for physical AI application deployment
Applications
The NVIDIA IGX T5000 module is ideal for a wide range of edge AI applications:
- Collaborative robots and humanoid robotics
- Autonomous mobile robots (AMRs) and automated guided vehicles (AGVs)
- Medical imaging systems and surgical robotics
- Industrial inspection and quality control
- Smart manufacturing and Industry 4.0
- Real-time video analytics and surveillance
- Autonomous vehicles and transportation systems
- Edge AI inference for generative AI models
What's in the Box
- NVIDIA IGX T5000 System-on-Module
- Documentation (varies by distributor)
Note: The IGX T5000 is a production module designed for integration into custom carrier boards. For development purposes, consider the NVIDIA IGX Thor Developer Kit Mini, which includes the T5000 module in a full chassis enclosure with safety MCU.
Full Specifications
Compute
| GPU Architecture |
NVIDIA Blackwell |
| CUDA Cores |
2,560 |
| Tensor Cores |
96 (5th Generation) |
| AI Performance |
Up to 2,070 FP4 TFLOPS / 1,035 FP8 TFLOPS |
| CPU |
14-core Arm Neoverse-V3AE 64-bit CPU |
| Safety Cores |
4× Cortex-R52 in Dual-Core Lock-Step (DCLS) mode |
Memory
| System Memory |
128 GB LPDDR5X |
| Memory Interface |
256-bit |
| Memory Bandwidth |
273 GB/s |
Video Processing
| Video Encode |
2× NVENC (8K30 / multiple 4K60 streams) |
| Video Decode |
2× NVDEC (8K30 / multiple 4K60 streams) |
| Vision Accelerator |
PVA v3.0 (Programmable Vision Accelerator) |
Camera & Sensor Interfaces
| MIPI CSI-2 |
16-lane interface (164 Gbps C-PHY v2.1 / 40 Gbps D-PHY v2.1) |
| Camera Streams |
Up to 32 simultaneous camera data streams with Virtual Channels |
| Holoscan Sensor Bridge |
Connectivity for up to 20 cameras |
Networking & Connectivity
| Network Support |
4× 25 GbE |
| High-Speed I/O |
PCIe Gen5, flexible I/O for custom carrier boards |
Power & Thermal
| Power Envelope |
Configurable 40–130 W |
Functional Safety
| Safety Standards |
ISO 26262, IEC 61508 (targeting ASIL D / SIL 3) |
| Safety Features |
Functional Safety Island (FSI) on SoC, Hardware Security Module (HSM) |
Software & Support
| Software Stack |
NVIDIA IGX Software (NVIDIA AI Enterprise optional on T5000) |
| Supported Frameworks |
NVIDIA Isaac, Metropolis, Holoscan, Halos, NIM |
| Lifecycle Support |
Up to 10 years |
| Form Factor Compatibility |
Pin-compatible with NVIDIA Jetson T5000 |
Physical
| Form Factor |
System-on-Module (SoM) for custom carrier board integration |
Specifications are sourced from the manufacturer product page and may be updated.
