Foundation of GPU Programming

Explore the fundamentals of GPU programming in this comprehensive conference talk from Meeting C++ 2022. Delve into the world of heterogeneous computing systems, focusing on the power of Graphics Processing Units (GPUs) as general-purpose parallel processors. Gain insights into the software hierarchy of GPUs and discover the key hardware components that drive their exceptional performance. Learn about SIMD processing, DRAM bank-level parallelism, GPU software hierarchy, and memory hierarchy. If time permits, the talk also introduces the CUDA memory model and GPU programming model. Designed to boost confidence and comfort in using GPUs, this presentation aims to equip programmers with essential knowledge about the main components enabling GPU performance.