Genetic algorithms (GAs) are powerful solutions to optimization problems arising from manufacturing and logistic fields. It helps to find better solutions for complex and difficult cases, which are hard to be solved by using strict optimization methods. Accelerating parallel GAs with GPU computing have received significant attention from both practitioners and researchers, ever since the emergence of GPU-CPU heterogeneous architectures. Designing a parallel algorithm on GPU is different fundamentally from designing one on CPU. On CPU architecture, typically data or tasks are distributed across tens of threads or processes, while on GPU architecture, more than hundreds of thousands of threads run. In order to fully utilize the computing power of GPUs, the design approaches and implementation strategies of parallel GAs should be re-probed. In the chapter, a concise overview of parallel GAs on GPU is given from the perspective of GPU architecture. The concept of parallelism granularity is redefined, the aspect of data layout is discussed on how it will affect the kernel performance, and the hierarchy of threads is examined on how threads are organized in the grid and blocks to expose sufficient parallelism to GPU. Some future research is discussed. A hybrid parallel model, based on the feature of GPU architecture, is suggested to build up efficient parallel GAs for hyper-scale problems.
Part of the book: Industry 4.0