Multiobjective evaluation and optimization of CMT-bone on multiple CPU/GPU systems

Abstract Hybrid multi-core processors are poised to dominate the landscape of next generation computing. By integrating several types of cores on a single chip, designers are anticipating sustained performance growth while depending less on raw circuit speed and decreasing power requirements per unit of performance. In this paper, we study the performance/energy tradeoffs for a heterogeneous platform consisting of Intel multicore processor with multiple nVidia GPU co-processors. Using CMT-bone, a proxy-app designed at the University of Florida, we explore load balancing strategies for a variety of architectural combinations of CPUs and GPUs to optimize for performance, power and energy measurements.