Fetch Size Adaptation vs. Stream Buuer for Media Benchmarks 1 Contents 1 Introduction 1 2 Related Work 1 3 Impacts of Variable Fetch Sizes and Stream Buuer 2 4 Fetch Size Adaptation 6

In current cache designs cache size and line size are often xed and determined by the spatial and temporal locality of benchmarks used to evaluate the targeted processors Stream bu er and large fetch size are two techniques that exploit spatial locality to hide memory latency In this paper we compare the e ectiveness of stream bu er and fetch size on media benchmarks We have presented a dynamic prediction mechanism for fetch size adaptation Our results show that fetch size adaptation is as e ective as stream bu er and both can improve the performance of media benchmarks by up to Due to its simple hardware design fetch size adaptation is a viable alternative to stream bu er in hiding memory latency