Power-performance analysis of metaheuristic search algorithms on the GPU

This paper presents a power-performance analysis of three metaheuristic search algorithms on the GPU. We investigate the impact of varying thread configurations with constraints on the maximum allowable registers per thread. The experimental results reveal that thread geometry can have a significant impact on both the performance and power consumption of the studied codes. Generally, larger block sizes with a low constraint on the number of allocated registers provide the best results. A particularly interesting outcome of this study is the discovery of thread and grid dimensions which yield improvements in both performance and power consumption. These improvements come from better data locality across thread blocks and reduction in DRAM traffic.