Problem vectorizability and the market for vector supercomputers

摘要

Recent concern has been voiced over the loss of U.S. preeminence in supercomputer design and manufacture, and over the comparatively low usage of supercomputers in U.S. industry. This research investigates industry demand for supercomputing, and the supply characteristics of problems to which supercomputers are typically applied. The research investigates traditional problem vectorizability decision rules for selecting problems suitable for supercomputing, and suggests methods of improving the cost effectiveness of supercomputing. It shows that problem vectorizability decision rules cannot be assumed a priori to assure cost effective supercomputer performance. Rather, the benefits from supercomputing tend to be highly sensitive and situation specific. The failure to appropriately apply supercomputing to problems of commercial importance is a rational reaction to uncertainties associated with the cost effectiveness of supercomputing. This result provides insight into current industry trends in light of the framework presented for analysis, and suggests beneficial directions for supercomputing developments.