Contemporary Supercomputing: Opportunities for Science and Challenges for Computer Engineering

President Obama’s July 2015 Executive Order, which established a National Strategic Computing Initiative, ensures that the U.S. will make substantial investments in the development of exascale computing systems. While this opens many opportunities in science, construction of such systems calls for new approaches to software, mathematical algorithms and systems engineering.

In this lecture, Dr. Schulthess will show how recent developments in architecture have moved us away from traditional abstractions, forcing software development and mathematical algorithms to acknowledge the physical reality of computing systems. Data locality and asynchrony will be key to the effective use of exascale computing systems. Furthermore, the dusk of complementary metal-oxide semiconductor (CMOS) scaling is increasing the diversity of computer architectures. This is profoundly challenging to software development and systems engineering, but at the same time, it opens many new opportunities for science. A strategy to manage this software challenge will be discussed in terms of recent experiences in numerical weather predictions.