关键词：高效节能；拥抱内存墙；科学计算
摘 要：In this chapter, we demonstrated how we can embrace the memory wall to address the power wall for scientific computing. We proposed a software approach that provides the active-state power management of the CPU by means of DVFS and takes advantage of sublinear performance scaling in non-CPU activities. In detail, we presented a DVFS algorithm called the β-adaptation algorithm. This PMU-assisted, interval-based algorithm uses a compute-boundedness metric called β to capture the effect of sublinear performance scaling. By design, the algorithm minimizes its dependence on the PMU, which is essential for the portability of the algorithm.