Abstract 

Scientists and engineers rely on simulation to predict the behavior of devices under a variety of design condition Often, experiments are too expensive or hazardous to carry out or are incapable of providing detailed information that can enable critical insight into making better design decisions for increased efficiency and safety. Advanced science and technology simulation also often requires extreme computing resources that make it intractable by conventional computational tools.

I will present high-order methods, algorithms, and analysis on performance characteristics that impact the scalability of electromagnetic and fluid simulations on large-scale parallel computers including many-GPU and many-core architectures. Discussion includes an open-source simulation-software package NekCEM/Nek5000 featuring state-of-the-art, scalable, high-order algorithms that are fast and efficient on platforms ranging from laptops to the worlds fastest computers for a wide range of scientific applications including electromagnetics, fluid flow, thermal convection, combustion, and magnetohydrodynamics.