NVIDIA and partners build AI-enabled fusion reactor digital twin

NVIDIA, General Atomics and international research partners announced a high-fidelity, AI-enabled digital twin of the DIII-D fusion reactor at the NVIDIA GTC Washington, D.C. conference.

• Main announcement/action: Built an interactive AI-enabled digital twin of the US DOE DIII-D National Fusion Facility that fuses sensor data, physics-based simulations, engineering models and AI surrogate models; trained three surrogate models (EFIT, CAKE, ION ORB) at scale using Polaris (ALCF) and Perlmutter (NERSC) and running on NVIDIA GPUs, Omniverse, CUDA-X, RTX PRO Servers and DGX Spark. The twin is synchronized with the physical DIII-D and available to an international team of ~700 scientists from 100 organizations, enabling simulations to move from weeks to seconds.
• Background and supporting details: Technical support and contributions from San Diego Supercomputer Center (UC San Diego), Argonne National Laboratory (ALCF) and Lawrence Berkeley National Laboratory (NERSC); this work is part of General Atomics’ research at the US DOE DIII-D National Fusion Facility and was announced at NVIDIA GTC Washington, D.C..
  • Event: NVIDIA GTC Washington, D.C.
    • Date: not specified in article
    • Time: not specified in article
    • Location: Washington, D.C.
    • Agenda/subject: Announcement of AI-enabled digital twin for fusion research