Performance Analysis and Tuning for General Purpose Graphics Processing Units (GPGPU)

The book "Performance Analysis and Tuning for General Purpose Graphics Processing Units (GPGPU)" offers a comprehensive introduction to the architecture and programming of GPGPUs, which have established themselves as a significant class of parallel processing systems. This technology is applied in a variety of computers, from high-performance computing systems to mobile platforms. The work highlights the differences between GPGPUs and traditional multicore systems, particularly regarding hardware multithreading capabilities, memory capacities, and bandwidth. It provides an overview of current GPGPU architectures and discusses the principles of earlier parallel platforms. Additionally, it covers optimization strategies ranging from high-level algorithms to low-level instruction optimizations. A practical example, the n-body particle simulation using the fast multipole method, is used as a case study. The book concludes with a brief overview of modern tools and techniques for performance analysis of GPUs.