Parallel Programming

Chapter 3

Modern Architectures

Abstract

In the past decades, the increased single-core performance of every released processor generation has been the main contributor towards runtime improvements of applications. Nowadays, single-threaded CPU performance is stagnating and performance improvements of modern CPUs can mainly be attributed to an increasing number of cores. Nevertheless, taking full advantages of modern architectures requires in many cases not only sophisticated parallel algorithm design but also knowledge of modern architectural features. An important example is the memory system. Modern microprocessors can process data at a much faster rate than reading data from main memory – known as the von Neumann bottleneck. As a consequence, ...