ABSTRACT
The unofficial race for the first petaflops computer is on. MDGRAPE-3, a system with hardware specialized for molecular dynamics simulations was announced in June 2006 as the first system with a theoretical peak performance at this level [12]. An informal survey at the 2006 Dagstuhl workshop on Petascale Algorithms and Application showed that the first petaflop Linpack performance is widely expected by 2008. At the same time the DARPA High Productivity Computing Systems (HPCS) program leads the charge to develop more productive petascale systems, which are easier to program and should achieve higher effective performance levels. In this chapter we are looking at the general question of how to evaluate performance and productivity of current and future systems.
