At the GTC conference in Washington, D.C., NVIDIA unveiled NVIDIA IGX Thor, a new platform designed for industrial and medical edge computing. Designed to bring real-time physical AI capabilities directly to the edge, this platform integrates high-speed sensor processing, enterprise-grade reliability, and functional safety into a compact desktop-sized module.
NVIDIA emphasizes that compared to the previous generation NVIDIA IGX Orin, IGX Thor's AI computing performance is improved by up to 8 times, enabling developers to build faster, safer, and smarter systems, accelerating the transformation of AI from the digital world to the physical world.
Core architecture: Blackwell iGPU + dGPU, 5581 FP4 TFLOPS computing performance
To overcome the speed and capacity limitations of traditional edge systems in processing real-time data and securely running multiple generative AI models, IGX Thor incorporates powerful hardware specifications:
• Dual Blackwell GPUs: The platform has two built-in NVIDIA Blackwell architecture GPUs—an integrated GPU (iGPU) and a discrete GPU (dGPU).
• High AI computing power: Provides up to 5,581 FP4 TFLOPS of AI computing performance.
• High-speed connection: Supports 400 GbE network connectivity.
NVIDIA pointed out that compared to IGX Orin, IGX Thor's AI computing power on iGPU has increased by 8 times, dGPU has increased by 2.5 times, and connectivity has increased by 2 times, enabling it to run large language models (LLM) and visual language models (VLM) more smoothly on the edge.
Industrial-grade design: 10-year lifecycle, integrated Halos functional safety
Designed for the demanding needs of industrial and medical environments, the IGX Thor platform offers the following features:
• Industrial-grade platform: Provides a 10-year product lifecycle and long-term support for the NVIDIA AI software stack.
• Software support: Runs the NVIDIA AI Enterprise software stack, including NVIDIA NIM microservices (accelerating cloud-to-edge physical AI application development), NVIDIA Isaac (robotics), NVIDIA Metropolis (vision AI), and NVIDIA Holoscan (sensor processing).
• Integrated functional safety: Integrating elements from the NVIDIA Halos full-stack safety system, it embeds functional safety into robotics, medical, and industrial AI systems, leveraging onboard and infrastructure sensors to ensure safe human-machine collaboration.
Widely adopted by the ecosystem: Hitachi Rail, CMR Surgical, etc.
NVIDIA announced that several global leaders in industrial, robotics, medical, and aerospace will be early adopters of IGX Thor:
• Industrial/Robotics: Hitachi Rail (for railway predictive maintenance and autonomous inspection), Maven Robotics (for next-generation industrial robots), and the SETI Institute.
• Aerospace: Joby Aviation (developing eVTOL-related applications).
• Medical/Care: Diligent Robotics, EndoQuest Robotics, and CMR Surgical are evaluating the use of the IGX Thor in their surgical robotic systems to enable real-time analysis and adaptive decision-making.
In addition, an ecosystem of equipment manufacturers, including Advantech, ADLINK, ASRock Rack, Inventec, NexCOBOT, Onyx, and YUAN, including several Taiwanese manufacturers, will also provide a series of IGX Thor-based edge servers, customized carrier boards, design services, cameras, sensors, and AI and system software to accelerate solution development.
Synchronously expand Omniverse Blueprint to accelerate the development of factory digital twins and robotics
In a separate presentation at GTC DC, NVIDIA also announced the expansion of its “Mega” NVIDIA Omniverse Blueprint to include technology for designing and simulating factory-scale digital twins.
• Siemens becomes the first digital twin software developer to support this blueprint (Siemens Xcelerator platform).
• FANUC and Foxconn Fii (Hongbai) became the first manufacturers to support OpenUSD digital twin models of their robots. Jensen Huang demonstrated how Foxconn used Omniverse to design its NVIDIA AI server manufacturing plant in Houston, Texas.
• U.S. manufacturing leaders such as Belden, Caterpillar, Lucid Motors, Toyota, TSMC, and Wistron are using Omniverse to create digital twins of their factories and accelerate AI-driven manufacturing processes.
• Robotics companies such as Agility Robotics, Amazon Robotics, Figure, and Skild AI are using NVIDIA’s three-computer architecture (including Jetson AGX Thor, IGX Thor, etc.) to build collaborative robots.
Listing Information
The NVIDIA IGX Thor platform will be available as two production-ready systems: the NVIDIA IGX T5000 module and the NVIDIA IGX T7000 development board kit. Both the production systems and development kits are expected to be available starting in December.
