Improving Effectiveness of Parallel Processing with Smart Cache Utilization for Energy Saving and Temperature Reduction In Chip Multiprocessors

Abstract

As the need for high computing is ever-increasing, chip multiprocessors have emerged as a promising solution for parallel and distributed processing. The number of execution units are increasing on a single chip for concurrent operations. Though parallel processing gives the fast execution, it comes with the drawback of high leakage energy and thereby chip temperature increase.With technology scaling, this problem has become more severe and causes performance degradation and chip malfunctioning.It even challenges the reliability of chip functioning and its safety. Chip multiprocessors come along with large on-chip caches.In chip multiprocessors,cache memory consumes an average of 30% of the overall chip energy. When a chip multiprocessor is used for any application, all the cache banks are not used all the time. Banks that are not used by any processor can be utilized as thermal buffers to dissipate energy and reduce the temperature.The on-chip cache can be effectively used with the proper distribution of the workloads. We, in our research, propose a technique to use cache banks to control the increase in chip temperature. With this method, we achieved up to 29% energy savings and up to 5.3Kelvin LLC temperature reduction. Both the techniques effectively use Last Level Cache (LLC) to reduce temperature and save energy.