"GPU Computing" is a definitive, hands-on guide designed to transform you from a novice into a proficient practitioner of parallel programming on modern Graphics Processing Units (GPUs). This book is built on a direct and pragmatic philosophy, meticulously structured to provide the skills required by top-tier academic institutions and the high-performance computing industry worldwide.
Philosophy: From Theory to Implementation
The core philosophy of GPU Computing is "learning by doing." The field of parallel programming is inherently practical. While understanding the theory of parallel architectures is important, true mastery comes from the hands-on process of designing, writing, debugging, and optimizing parallel code. This book is engineered to facilitate that active learning process.
Key Features
1. Strictly Practical Focus: Heavily prioritizes implementation, coding, and application development over abstract theory.
2. Beginner to Advanced Path: The structure supports those with no parallel programming experience while also providing advanced chapters on optimization, profiling, and multi-GPU systems for more experienced learners.
3. CUDA-Oriented: Focuses on NVIDIA's CUDA, the most mature and widely-used platform for general-purpose GPU computing in both industry and academia.
5. Ten-Chapter Structure: A concise and focused structure that covers all essential topics without unnecessary filler, making it ideal for a semester-long course.
6. Complete Capstone Project: Includes a live, do-it-yourself project with fully explained, working code to provide a portfolio-worthy development experience.
7. Latest and Updated Content: Covers modern GPU architectures, relevant CUDA features, and current best practices in the field.
Key Takeaways
Upon completing this book, you will be able to:
1. Understand GPU Architecture: Explain the fundamental design of a modern GPU, including its streaming multiprocessors, cores, and memory hierarchy.
2. Write and Launch CUDA Kernels: Develop parallel programs in CUDA C++ and execute them on the GPU.
3. Manage GPU Memory: Efficiently allocate and transfer data between the CPU (host) and GPU (device) and utilize different memory spaces (global, shared, constant) for optimal performance.
4. Implement Parallel Algorithms: Convert sequential algorithms into their parallel counterparts using common patterns like reduction, scan, and parallel sorting.
5. Optimize and Debug Code: Use profiling tools to identify performance bottlenecks and apply standard optimization techniques. Debug complex parallel code effectively.
6. Utilize CUDA Libraries: Leverage powerful, pre-built NVIDIA libraries (e.g., cuBLAS, cuFFT) to accelerate common computational tasks.
7. Build a Complete GPU-Accelerated Application: Design, implement, and deploy a functional application that effectively harnesses the power of the GPU from start to finish.
Disclaimer: Earnest request from the Author.
Kindly go through the table of contents and refer kindle edition for a glance on the related contents.
