Poster: Three Steps to Model Power-Performance Efficiency for Emergent GPU-Based Parallel Systems

According to the US DOE, there is a need for 1000-fold performance increase with only a 10-fold power increase to achieve the goal of a sustained petaflop within 20 Megawatts by 2019. Despite the top-ranked energy efficiency supercomputer at about 2,000 MFLOPS/watt, to meet the DOE goal of 50,000 MFLOPS/watt will require significant innovation in power and performance efficiency. In order to help application and system designers find power-performance bottlenecks and their causes, a need for accurate models that balance abstraction and detail on emergent GPU architectures.