IGX Thor delivers substantial AI performance gains at the edge, including up to 8x higher AI compute on the integrated GPU and up to 2x higher with discrete GPU acceleration versus IGX Orin, along with significantly increased networking bandwidth for real-time workloads.

NVIDIA IGX Thor is an enterprise-grade edge AI platform designed for industrial and medical environments, combining high AI performance with rugged hardware, extended lifecycle support, and comprehensive functional safety features to ensure deterministic and safe edge operation.

IGX Thor emphasizes real-time determinism for edge applications through a preemptible real-time Linux kernel, multi-instance GPU for isolated safety-critical tasks, programmable accelerators for vision and real-time inference, and GPU Direct RDMA for direct sensor-to-GPU data flow.

The platform is built rugged for harsh industrial environments, featuring industrial-grade components, rugged chassis options, a long 10-year lifecycle, and extensive I/O to integrate with legacy PLCs, sensors, and control networks.

NVIDIA AI Enterprise software stack provides production-ready, enterprise-grade support with long-term lifecycle commitments (up to a decade) and a stable LTS AI framework/SDK path.

NVIDIA’s Halos AI Systems Inspection Lab, ANAB-accredited, offers independent safety assessments to support customer certification and strengthen industrial deployment readiness.

IGX Thor Developer Kit and Developer Kit Mini are available now through global distributors, enabling immediate prototyping and early deployments.

The transition from Jetson Thor to IGX Thor is designed to be seamless, with compatible pinouts and form factors, aligned software stacks, and JetPack-on-IGX offerings for deeper customization.

Networking improvements include dual 200 GbE connections powered by ConnectX-7 SmartNICs, enabling GPU Direct RDMA to bypass the CPU for sensor data and reduce latency in real-time applications.

High-speed connectivity is core, with dual 200 GbE networking enabling RDMA and direct GPU memory access for low-latency sensor data processing and scalable multisensor pipelines via the Holoscan Sensor Bridge.

Real-world use cases span high-concurrency edge workloads in smart retail, industrial safety, and building automation, including safety-critical robotics and medical applications, supported by data on performance, safety, and reliability.

The launch aligns with NVIDIA’s broader push into physical AI applications across industries, expanding beyond data-center GPUs into robotics, automation, and industrial edge devices.