GPUs have recently burst onto the scientific computing scene as an innovative technology that has demonstrated substantial performance and energy efficiency improvements for the numerous scientific applications. These initial applications were often pioneered by early adopters, who went to great effort to make use of GPUs. More recently, the critical question facing this technology is whether it can become pervasive across the multiple, diverse algorithms in scientific computing, and useful to a broad range of users, not only the early adopters. A key barrier to this wider adoption is software development: writing and optimizing massively parallel CUDA code, using new performance and correctness tools, leveraging libraries, and understanding the GPU architecture.

Part of this challenge will be solved by experts sharing their knowledge and methodology with other users through books, tutorials, and collaboration. CUDA Application Design and Development is one such book. In this book, the author provides clear, detailed explanations of implementing important algorithms, such as algorithms in quantum chemistry, machine learning, and computer vision methods, on GPUs. Not only does the book describe the methodologies that underpin GPU programming, but it describes how to recast algorithms to maximize the benefit of GPU architectures. In addition, the book provides many case studies, which are used to explain and reinforce important GPU concepts like CUDA threads, the GPU memory hierarchy, and scalability across multiple GPUs including an MPI example demonstrated near-linear scaling to 500 GPUs.

Lastly, no programming language stands alone. Arguably, for any language to be successful, it must be surrounded by an ecosystem of powerful compilers, performance and correctness tools, and optimized libraries. These pragmatic aspects of software development are often the most important factor to developing applications quickly. CUDA Application Design and Development does not disappoint in this area, as it devotes multiple chapters to describing how to use CUDA compilers, debuggers, performance profilers, libraries, and interoperability with other languages.

I have enjoyed learning from this book, and I am certain you will also.

20 September 2011