NVIDIA IGX Thor
Product · Hardware / device · Engineering & development by NVIDIA
The next-generation, enterprise-ready AI platform for physical AI.
NVIDIA IGX Thor is an industrial-grade edge AI platform for safety-critical machines in medical and industrial sectors. It delivers up to 5581 TFLOPS of AI performance with Blackwell architecture and 10 years of support.
Category: Edge Computing
- Performance
- Up to 5581 TFLOPS (FP4-Sparse) AI performance
- Interfaces and ports
- NVIDIA ConnectX-7 networking, dGPU support, BMC support
- AI Performance (FP4-Sparse)
- Up to 5581 TFLOPS
- GPU Architecture
- NVIDIA Blackwell with 5th-gen Tensor Cores
- Safety Processor
- Dedicated Functional Safety Island
- Safety Standards
- Designed for ISO 26262 and IEC 61508
- Up to 5581 TFLOPS FP4 AI performance
- NVIDIA Blackwell GPU architecture with 5th-gen Tensor Cores
- Up to 10 years of enterprise software support
- Integrated NVIDIA ConnectX-7 high-speed networking
Overview
NVIDIA IGX Thor is an industrial-grade edge AI platform designed for autonomous, safety-critical machines in medical, industrial, and robotics environments. It combines enterprise-level hardware with the NVIDIA Blackwell GPU architecture to deliver high-performance generative reasoning and multimodal sensor processing at the edge. The platform features a dedicated Functional Safety Island and runs NVIDIA Halos OS, providing a foundation for human-machine collaboration that meets ISO 26262 and IEC 61508 standards. IGX Thor delivers up to 5581 FP4 TFLOPS of AI compute, enabling the execution of multiple generative AI models simultaneously. It is available in various forms, including developer kits and production-ready systems, and comes with up to 10 years of enterprise support. This full-stack solution integrates NVIDIA AI Enterprise software to ensure mission-critical performance and security for next-generation physical AI applications.
Compute and Safety Architecture
- 5581 TFLOPS AI Performance: The platform delivers up to 5581 FP4 TFLOPS of AI compute to run multiple generative AI models simultaneously at the edge.
- Blackwell GPU Architecture: NVIDIA IGX Thor is powered by the NVIDIA Blackwell GPU architecture with fifth-generation Tensor Cores and Multi-Instance GPU (MIG) support.
- Functional Safety Island: The system includes a dedicated Functional Safety Island with an independent safety processor to isolate safety-critical workloads.
- Safety Standards Compliance: The platform is designed to meet rigorous functional safety standards including ISO 26262 and IEC 61508 for industrial and automotive environments.
Software and AI Ecosystem
- NVIDIA Halos OS provides a real-time functional safety foundation adapted from proven autonomous vehicle safety stacks for robotics.
- NVIDIA AI Enterprise Integration: The platform includes NVIDIA AI Enterprise software to ensure mission-critical performance and security across the entire stack.
- Cloud-Native Edge Deployment: IGX Thor supports cloud-native NVIDIA NIM microservices to bring generative AI models like LLMs and VLMs to edge environments.
- Industry-Specific SDKs: The software stack supports specialized frameworks including NVIDIA Isaac for robotics, Metropolis for visual AI, and Holoscan for sensor processing.
Connectivity and Lifecycle
- High-Bandwidth Sensor Processing: The platform provides high-bandwidth sensor processing capabilities for real-time applications in medical and industrial settings.
- Advanced Networking: IGX Thor features NVIDIA ConnectX-7 for high-speed networking and 2x better connectivity compared to previous generations.
- 10-Year Enterprise Support: The platform is offered with up to 10 years of enterprise support to maintain long-term reliability in challenging conditions.
- Industrial-Grade Reliability: The hardware is purpose-built for human-machine collaboration in dynamic environments like factories and operating rooms.
Use cases
- Medical device manufacturers use IGX Thor to power real-time sensor processing and AI reasoning for next-generation surgical and diagnostic equipment.
- Industrial robotics companies deploy IGX Thor to enable autonomous robots to operate safely alongside humans in dynamic factory environments.
- Transportation providers integrate IGX Thor into autonomous machines to handle complex generative reasoning and multimodal sensor data at the edge.
- Smart factory operators use the platform to run multiple generative AI models for real-time industrial inspection and process optimization.
Specifications
- Performance
- Up to 5581 TFLOPS (FP4-Sparse) AI performance
- Interfaces and ports
- NVIDIA ConnectX-7 networking, dGPU support, BMC support
- Operating conditions
- Industrial-grade for robotics, medical, and industrial environments
- Sizes and variants
- IGX Thor Developer Kit, IGX Thor Developer Kit Mini, IGX T7000 MicroATX, IGX T5000 System-on-Module
- AI Performance (FP4-Sparse)
- Up to 5581 TFLOPS
- GPU Architecture
- NVIDIA Blackwell with 5th-gen Tensor Cores
- Safety Processor
- Dedicated Functional Safety Island
- Safety Standards
- Designed for ISO 26262 and IEC 61508
- Networking
- NVIDIA ConnectX-7 integrated
- Support Lifecycle
- Up to 10 years
- iGPU Configuration
- 2560-core NVIDIA Blackwell with 10 TPCs
- dGPU Option
- NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition
How it works
- Acquire Developer Kit: Developers can purchase IGX Thor Developer Kits or Developer Kit Mini systems from trusted NVIDIA distributors to begin prototyping physical AI applications.
- Develop with NVIDIA AI Stack: Utilize NVIDIA AI Enterprise, Isaac for robotics, or Holoscan for medical sensor processing to build and test generative AI models at the edge.
- Configure Functional Safety: Implement safety-critical workloads using the Functional Safety Island and NVIDIA Halos OS to meet industrial and medical safety standards.
- Deploy Production Systems: Transition to production using IGX Thor production systems or System-on-Modules (SoM) for custom carrier board integrations in industrial machines.
FAQ
How does IGX Thor performance compare to previous generations?
The platform delivers up to 5581 FP4 TFLOPS of AI compute, which is up to 8x higher AI compute on the iGPU compared to the previous IGX Orin generation. It also offers 2x better connectivity and 2.5x higher AI compute on the dGPU.
What safety features are built into IGX Thor?
IGX Thor includes a dedicated Functional Safety Island with an independent safety processor and runs NVIDIA Halos OS. It is designed to meet ISO 26262 and IEC 61508 standards to ensure safe operation in dynamic environments.
What is the support lifecycle for IGX Thor?
NVIDIA provides up to 10 years of enterprise support for the IGX platform, ensuring long-term stability, security, and performance for mission-critical industrial and medical applications.
Added by AntSpace from public sources (nvidia.com).