NVIDIA PhysicsNeMo

NVIDIA PhysicsNeMo is an open-source Python framework for engineers and researchers building AI models of physical systems. It combines physics constraints with simulation and observed data to train surrogate models that can predict behavior in near real time. The framework uses PyTorch and carries the Apache 2.0 license.

Its model families cover different kinds of scientific data. Fourier neural operators work with regular grids, MeshGraphNet handles mesh-based dynamics, and DoMINO predicts surface and volume fields for automotive aerodynamics. Physics-informed neural networks and diffusion models support physics constraints, generative tasks and inverse problems. Components also work with point clouds, graphs and nested physical fields.

You can reuse individual layers and numerical operators within an existing PyTorch workflow, or build on complete models and training recipes. Reference applications cover fluid flow, structural mechanics and electromagnetics, with further examples for weather forecasting, cardiovascular flow and data-center thermal design. Pretrained models are available through NVIDIA NGC.

Training runs on NVIDIA GPUs, from a single GPU to multi-node clusters. Domain parallelism can split one high-resolution sample across GPUs when the sample itself is too large for one device. NVIDIA provides development containers; its separate External Aerodynamics NIM packages a pretrained model for cloud deployment, while PhysicsNeMo includes the recipe for further customization.