Monday, June 14, 2010
Tensilica - Berkeley Labs - Colorado Exaflop System Design
Next generation chips and networking available before 2015 would push the system to exaflop performance. Very narrow range of applications in mind. An application-driven architecture does not necessitate a special-purpose machine, nor does it require exotic technology.Designed for highly parallel applications. With software tuning, Green Flash architetcure can reduce a per-core computational requirement of 3.5 gigaflops to a more feasible 0.5 gigaflops.