부산대학교 도서관

GPU computing has spread its capacity over most of the scientific computing areas. Soft computing is aware of the potential of this computing architecture. In order to achieve high performance, practitioners have to deal with the particularities associated with the porting of the problem to the specifications of the GPU card; and specially with an efficient management of the on-board and on-chip memories. In this work, diverse optimized data layouts are analysed when evaluating a large population of high-dimensional individuals for Rosenbrock function. As a consequence of this study, a statement about the most favourable data layout for this kind of evaluation is presented.