SC19 Proceedings

The International Conference for High Performance Computing, Networking, Storage, and Analysis

SW_GROMACS: Accelerate GROMACS on SUNWAY TaihuLight


Authors: Tingjian Zhang (Shandong University; National Supercomputing Center, Wuxi), Yuxuan Li (Tsinghua University, China; National Supercomputing Center, Wuxi), Ping Gao (Shandong University; National Supercomputing Center, Wuxi), Qi Shao (Shandong University; National Supercomputing Center, Wuxi), Mingshan Shao (Shandong University; National Supercomputing Center, Wuxi), Meng Zhang (Shandong University; National Supercomputing Center, Wuxi), Jinxiao Zhang (Shandong University), Xiaohui Duan (Shandong University; National Supercomputing Center, Wuxi), Zhao Liu (Tsinghua University, China; National Supercomputing Center, Wuxi), Lin Gan (Tsinghua University, China; National Supercomputing Center, Wuxi), Haohuan Fu (Tsinghua University, China; National Supercomputing Center, Wuxi), Wei Xue (Tsinghua University, China; National Supercomputing Center, Wuxi), Weiguo Liu (Shandong University; National Supercomputing Center, Wuxi), Guangwen Yang (Tsinghua University, China; National Supercomputing Center, Wuxi)

Abstract: GROMACS is one of the most popular Molecular Dynamic (MD) applications and is widely used in the field of chemical and biomolecular system study. Similar to other MD applications, it is computationally demanding. Therefore, many high-performance platforms have been employed to accelerate it, such as the KNL, Cell Processor, GPU and so on. As the third fastest supercomputer in the world, the TaihuLight contains 40,960 SW26010 processors, and every processor is a typical many-core processor. To make full use of the superior computation ability of the TaihuLight, we port the GROMACS to the SW26010. Finally, we not only achieved the 60× speedup in the hot spot kernel but also came up with a new useful strategy to accelerate the short-range interaction computation. This strategy could deal with the update conflict in the many-core processor acceleration without much performance hit. Besides SW26010, it could also be used in many other processors.


Presentation: file


Back to Technical Papers Archive Listing