Sunway TaihuLight: Designing and Tuning Scientific Applications at the Scale of 10-Million Cores

Abstract:

The Sunway TaihuLight supercomputer is the world's first system with a peak performance greater than 100 PFlops, and a parallel scale of more than 10 million cores.

In contrast with other existing heterogeneous supercomputers, which include both CPU processors and PCIe-connected many-core accelerators (NVIDIA GPU or Intel MIC), the computing power of TaihuLight is provided by a homegrown many-core SW26010 CPU that includes both the management processing elements (MPEs) and computing processing elements (CPEs) in one chip.

This talk will report our most recent progress on utilising the computing resource of more than 10 million cores for traditional scientific simulation applications, such as climate modeling, earthquake simulation, n-body, CFD, and big data analytic applications, such as remote sensing data processing, and Sunway-based deeplearning framework. Based on that, we will try to provide a picture of the major challenges as we see when mapping either new algorithms or existing softwares onto the Sunway TaihuLight supercomputer, and the major advantages brought by this machine. 

Speaker bio:

Dr Haohuan Fu is Deputy Director of the National Supercomputing Center in Wuxi, China, and leads the research and development efforts on Sunway TaihuLight, the current fastest supercomputer in the world. He is also an associate professor in the Ministry of Education Key Laboratory for Earth System Modeling, and Department of Earth System Science in Tsinghua University, where he leads the research group of High Performance Geo-Computing (HPGC).

Since joining Tsinghua in 2011, Dr Fu has been working towards the goal of providing both the most efficient simulation platforms and the most intelligent data management and analysis platforms for geoscience applications. His research has, for example, led to efficient designs of atmospheric dynamic solvers for both Tianhe-1A, Tianhe-2, Sunway TaihuLight supercomputers, and the reconfigurable computing platforms. The work based on reconfigurable technology can be both faster and more energy efficient than the Tianhe-1A and Tianhe-2 supercomputer, leading to a publication selected as one of the 27 most Significant Papers of the FPL conference in 25 years (27 out of 1765). The work based on the Sunway TaihuLight supercomputer manages to scale a fully-implicit solver to over 10 million cores, which won the Gordon Bell Prize of SC16.

Dr Fu has a PhD in computing from Imperial College London.