PetaFLOPS scale computing systems, opportunities, and challenges

Abstract

Petaflops is a scale of computer performance equal to a million billion operations per second and is thousands of times more powerful than today's most powerful massively parallel processors. Although representative of a class of system far beyond feasibility with contemporary technology, petaflops computer architecture is the target of active investigation and the topic of government sponsored workshops. Realizing petaflops capability will depend on advances in device technology, architectural structures, and parallel algorithms for scientific and engineering applications. This paper describes the field of petaflops computing in terms of the evolution of enabling technologies, new architectures likely to deliver effective performance at the petaflops level, and examples of important applications that will be significantly advanced through the availability of future petaflops computers. A summary of the seminal findings of several recent workshops exploring this regime of computing systems will be presented in detail. Analysis of the critical technologies and their evolution will be used to support estimates of the likely timeframe in which such capability will become practical.