PNNL

Abstract

Seymour Cray had a “Let’s go to the moon” attitude when it came to building high-performance computers. His drive was to create architectures designed to solve the most challenging problems. Modern high-performance computer architects, however, seem to be focusing on building the largest floating-point-generation machines by using truckloads of commodity parts. Don’t get me wrong; current clusters can solve a class of problems that are untouchable by any other system in the world, including the supercomputers of yesteryear. Many of the world’s fastest clusters provide new insights into weather forecasting, our understanding of fundamental sciences and provide the ability to model our nuclear stockpiles. Let’s call this class of problem a “first-principles simulation” because the simulations are based on a fundamental physical understanding or model.