Johan Barthélemy (University of Wollongong, NVIDIA)

Title : Accelerating Simulations with NVIDIA Modulus

Abstract : High-fidelity simulations in science and engineering are computationally expensive and time-prohibitive for quick iterative use cases, from design analysis to optimization.

NVIDIA Modulus, the open source physics machine learning platform, turbocharges such use cases by building physics-based deep learning models that are order of magnitudes. faster than traditional methods and offer high-fidelity simulation results.  Once trained, the model can perform quick forward passes, making it ideal for applications that demand fast responses, such as real-time simulations of large complex systems.

Modulus is a offers a suite Physics-ML model architectures, including Fourier neural operators. It provides an end-to-end pipeline for training models, from geometry ingestion to training and inference, with explicit parameter specifications for a wide range of applications.  The framework is also integrated with NVIDIA Omniverse for enhanced visualization and is tailored for high-performance, leveraging technologies for multi-GPU computing and multinode scaling.

This seminar will present the key features, use cases, and performance aspects of NVIDIA Modulus, providing attendees with a understanding of its capabilities and applications.

The seminar will take place in Room S07 at the Faculty of Sciences.