In this talk, she will describe runtime performance maximization strategies for hybrid CPU-GPU applications under the power constrains.  Both analytical and machine learning (ML) models will be presented along with the choices enabling their deployment during the execution of a diverse set of applications.  She will show experimental results with an Exascale Computing project (ECP) application GAMESS, which is a widely used package for ab initio quantum chemistry calculations, developed at Iowa State University. 

Power and the subsequent energy consumption pose major challenges in designing large scale systems.  In the exascale computing epoch, power efficiency at both software and hardware levels has become imperative for reducing the operating costs and failure rates.  Maximizing performance-per-watt entails a judicious distribution of the power budget, including heterogeneous host-accelerator configurations which are ubiquitous in current exascale platforms.

About Dr. Sosonkina

Masha Sosonkina has received her Ph.D. degree in Computer Science and Applications from Virginia Tech.  Upon graduation, she held faculty appointments at the University of Minnesota Duluth.  Dr. Sosonkina was a senior scientist at the U.S. Department of Energy Ames National Laboratory and held adjunct faculty appointments at Iowa State University.  She is now a Professor in the Electrical and Computer Engineering Department at Old Dominion University.  Her research interests include high-performance computing, large-scale cyber-physical simulations, performance analysis, and energy savings.