NVIDIA PhysicsNeMo is a free, open-source deep-learning framework for building physics-ML models โ neural networks that learn to simulate physical systems orders of magnitude faster than traditional numerical solvers. It provides the building blocks for digital twins and engineering simulation across fluid dynamics, structural mechanics, electromagnetics and climate science.
PhysicsNeMo (formerly known as Modulus) is an open-source project on GitHub from NVIDIA, with about 3,400 stars and active development in 2026. Built on PyTorch, it includes physics-informed neural network (PINN) utilities, neural operator architectures such as Fourier Neural Operators, graph-based simulators like MeshGraphNet, and optimised data pipelines for large engineering datasets. Reference applications cover problems like flow simulation over airfoils and Darcy flow, and the framework is designed to scale from a workstation GPU to multi-GPU clusters with NVIDIA Warp integration for differentiable physics. It is a core part of NVIDIA’s Omniverse and digital-twin ecosystem for industrial simulation.
PhysicsNeMo is completely free and open source under the Apache-2.0 license. You run it on your own GPU hardware with no license fees.
It also provides utilities for geometry handling, mesh-based data loading and distributed checkpointing, plus curated example workflows that teams can adapt to proprietary engineering datasets. Documentation, tutorials and reference applications are maintained alongside the code by NVIDIA’s simulation team.
PhysicsNeMo is for simulation engineers, computational scientists and ML researchers who want to accelerate physics simulation with neural surrogates, build digital twins, or research physics-informed machine learning.