Factors Influencing the Design of Next-Generation

High-End Computing and Visualization Architectures

Today's methods of scientific and engineering investigation range from the oretical, experimental to computational science. In computational science, the classical approach has been modeling and simulation. The concern here is the growing gap between actual applications and peak compute performances. We believe one major solution to this growing performance gap is the new multi-paradigm computing architecture. It tightly integrates, what were previously, disparate computing architectures into a highly scalable single system and, thus, allows them to cooperate on the same data residing in scalable globally-addressable memory. Enabling scientists to focus on science, not computer science. Additionally, with globally-addressable memory growing to Terascale sizes, a plethora of new, huge-memory applications that profoundly improve scientific and engineering productivity will come on line. From these, may emerge a new branch of computational science called data intensive methods. It includes the traditional method of query, to the more abstract methods of inference and even interactive data exploration. The availability of such a powerful range of interactive methods, for operation on Terascale data sets, all residing in monolithic globally-addressable memory, is a novel combination that will not only facilitate intended discoveries but may also give rise to a new complement which I will call 'planned serendipity'. The latter will be of growing significance in intelligence, science and engineering. And as the amount of data generated by faster and more productive systems grows, visualization will increasingly become an essential tool. The recent advances in display and related technologies, could pave the way for revolutionary new ways of visual, interactive and collaborative communications.