Article Preview
TopSince 2012, Intel Corporation has launched a new generation of coprocessor products, which has been paid great attention in the field of high-performance computing.
Fast multi-pole method (FMM) has small computing complexity, widely used in the field of astrophysics, molecular dynamics and other fields of numerical simulation problems. The United States national high-performance computing center (NCSA) ported the fast multilevel method software (ExaFMM) to the MIC platform (Pavlovl, Andonov & Kremenliev, 2013). Fu et al. of the Chinese academy of sciences and Tsinghua University have developed the global atmospheric simulation application on the Tianhe-2, mainly solving the shallow water equation (SWE) (Fu, 2013). Through MPI parallelism between nodes and vectorization optimization, the weak extended to 8652 computing nodes. Park et al. developed an anti-projection synthetic aperture radar application on Endeavor heterogeneous array with two MIC nodes (Park et al., 2012). It used MPI parallel between nodes, multi-mic in each mode, and CPU-MIC collaborative calculation. Lai, Huang, Shi et al. simulated the geographic space application (ISODATA) on the Beacon heterogeneous system with four MIC in each single node (Lai, Huang, Shi et al., 2013), and extended up to 120 MIC with symmetric mode. Mario Hernandez et al. analyzed the operation of the scientific application of 3D finite difference algorithm in the Intel MIC architecture (Chai, 2014), studied the limiting factors of the performance of the algorithm from the aspects of scalability, affinity, block size and grid shape, and optimized the algorithm.