James Stewart, Sandia National Laboratories
The U.S. is on a path to achieve exascale computing by 2022, with early hardware systems scheduled to become available in 2021. These systems are characterized by heterogeneous hardware that includes GPU (Graphical Processing Unit) accelerators from a variety of vendors. This heterogeneous computing environment requires a rethinking of our computational mechanics software and analysis paradigms to realize these performance gains. In this presentation, we discuss how these new challenges are being addressed across a variety of areas, such as: hardware/software co-design, performance portability, solver concurrency, meshing and geometry, uncertainty quantification, and visualization. We will also discuss progress being made in various application areas including wind turbines, climate modeling, materials, and machine learning.