, Director, Mathematics and Computer Science Division; Argonne Distinguished Fellow, Argonne National Laboratory
HPC systems are growing more complex, with hierarchically configured manycore processors, GPUs, and other accelerators, together with a deep and complex memory hierarchy. Due to the increase in complexity, applications face significant challenges with exploiting various system features to achieve good energy efficiency. To address these challenges, it's important to understand the trade-offs among runtime and power from both the applications and systems perspectives. We'll discuss our work in this area, and lessons we learned from exploring different applications across different systems.