Evaluation of leading scalar and vector architectures for scientific computations

Simon, Horst D. ; Oliker, Leonid ; Canning, Andrew ; Carter, Jonathan ; Ethier, Stephane ; Shalf, John

Lawrence Berkeley National Laboratory

关键词: Vectors;    Scalars;    Simulation;    42 Engineering;    Simulators;   

DOI  :  10.2172/833953
RP-ID  :  LBNL--55291
RP-ID  :  AC03-76SF00098
RP-ID  :  833953

美国|英语

来源: UNT Digital Library

【 摘 要 】

The growing gap between sustained and peak performance for scientific applications is a well-known problem in high performance computing. The recent development of parallel vector systems offers the potential to reduce this gap for many computational science codes and deliver a substantial increase in computing capabilities. This project examines the performance of the cacheless vector Earth Simulator (ES) and compares it to superscalar cache-based IBM Power3 system. Results demonstrate that the ES is significantly faster than the Power3 architecture, highlighting the tremendous potential advantage of the ES for numerical simulation. However, vectorization of a particle-in-cell application (GTC) greatly increased the memory footprint preventing loop-level parallelism and limiting scalability potential.