On November 17th, a Chinese team won the 2017 ACM Gordon Bell Prize at the Supercomputing Conference (SC2017) held in Denver, USA. The winning research was entitled “18.9-Pflops Nonlinear Earthquake Simulation on Sunway TaihuLight: Enabling Depiction of 18-Hz and 8-Meter Scenarios", led by Associate Professor Fu Haohuan in the Department of Earth System Science, Tsinghua University, with collaborators from six institutes. After winning this award for the first time in 2016, Chinese scientists achieved a consecutive winning of this outstanding award.
Founded in 1987, the Gordon Bell Prize is known as the most significant award in high-performance computing (HPC). It was presented annually by the Association for Computing Machinery (ACM) at the top conference in the field of supercomputing in November to reward cutting-edge parallel computing research, especially the outstanding achievements of innovative applications based on high-performance computing facilities.
Besides the work that won the Gordon Bell Prize, another work led by Fu Haohuan, “Redesigning CAM-SE for Petascale Performance,” was also selected as one of the three finalists of the Gordon Bell Prize. Together with the work that won last year on Nov 25th, the Sunway TaihuLight supercomputer has become the incubator of the Gordon Bell Prize.
“Four consecutive championships", highlighting the absolute strength of the hardware
Located at the National Supercomputer Center in Wuxi, Jiangsu, China, Sunway TaihuLight is a Chinese homegrown supercomputer operated by Tsinghua University with over 10.5 M heterogeneous cores. This machine has a peak performance of 125 PFlops and a sustained performance of 93 PFlops.
In the latest release of the Top 500 List (November 13, 2017) that ranks the world’s most powerful supercomputing facilities, Sunway TaihuLight ranked No.1 for the fourth consecutive time. China surpassed the United States once again in the total number of supercomputers of the TOP 500 List, highlighting China’s leading position in the world for supercomputer hardware.
Multidisciplinary efforts, to create Sunway software ecosystem
When the research team from Tsinghua first arrived at the National Supercomputing Center in Wuxi in the summer of 2015, the Gordon Bell Prize seemed to be an unreachable dream.
The complete homemade processor design of Sunway TaihuLight has brought significant improvement in both computing performance and power efficiency. However, its unique hardware architecture also means that all major scientific and engineering computing software cannot be used on this supercomputer directly. The software ecosystem for the homemade hardware was very limited.
The limited software ecosystem brings challenges but also opportunities for the application software development team for Sunway architecture. In July, 2015, the hottest season of Wuxi, over 30 professors and students from Tsinghua university, Beijing Normal University, and the First Institute of Oceanography of National Bureau of Oceanography, started to work with the migrations of millions lines of code in the climate models to this homegrown system. A challenging journey toward developing an ecosystem for the Sunway architecture has begun since then.
After 3 months of intensive research and development, the students majored in computer science became experts of atmospheric and ocean simulation, while the students majored in geosciences also gradually became professional programmers of Sunway TaihuLight. Finally they created the miracle of migrating millions lines of code and refactoring hundreds of thousands lines of code. In hindsight, the training of a group of young multi-disciplinary talents during the three months was the key to the successful application by the multi-institute team on Sunway TaihuLight.
After more than one year of application development, with the joint efforts of many universities and research institutions in China, a reasonable application software ecology is formed, with over ten science and engineering application software packages that can scale to the full machine with over 10 million cores.
A part of Tsinghua team continued to form a joint team with Beijing Normal University and Shandong University, and continued to work on the refactoring and redesign of the atmospheric model for Sunway TaihuLight. Meanwhile, another part of the Tsinghua team formed a joint team with Shandong University, Southern University of Science and Technology, and University of Science and Technology of China, and accomplished the earthquake simulation software framework on Sunway TaihuLight within three months.
Innovative methods with multiple world-leading applications
The research project named “18.9-Pflops Nonlinear Earthquake Simulation on Sunway TaihuLight: Enabling Depiction of 18-Hz and 8-Meter Scenarios” was awarded the 2017 ACM Gordon Bell Prize. This was a joint project by the Department of Earth System Science and the Department of Computer Science and Technology, Tsinghua University, and researchers from Shandong University, the Southern University of Science and Technology, the University of Science and Technology of China, the National Research Center of Parallel Computer Engineering and Technology, China, and the National Supercomputing Center, Wuxi, China.
Earthquake simulation tools can reproduce and simulate the process of earthquakes. It is an important method for scientists to understand the principles of geology, earthquake occurrence and propagation, and plays a significant role in reducing and preventing the huge losses caused by earthquake disasters. Seismic simulation tools can also be combined with other technologies to rationally design the infrastructure for high-seismic areas, in order to enhance the safety of urban planning.
Based on the powerful computing power of “Sunway TaihuLight”, the research team succeeded in designing and implementing a highly scalable non-linear large-scale earthquake simulation tool. This tool achieves a performance of up to 18.9-PFlops for nonlinear earthquake simulation, and for the first time in the world, achieves such large-scale, high-resolution and high-frequency non-linear plasticity earthquake simulations. It is also the first time to achieve a highly accurate simulation of the Tangshan earthquake (M7.8, 1976), which can help to better understand the impact caused by the Tangshan earthquake and provide insights and reference for future earthquake research.
In addition, the other finalist work, also led by Dr. Fu Haohuan, “Redesigning CAM-SE for Peta-Scale Performance”, applied a full set of optimization methods to refactor and redesign the classical atmospheric model CAM-SE for Sunway TaihuLight. The redesigned version can use up to millions of cores, and achieves a simulation speed of 3.4 SYPD (simulated years per day). The redesigned dynamic core can support a high resolution of 750m, and achieves a computing performance of 3.3 Pflops. Using the redesigned CAM software, the team achieved an accurate simulation of the entire lifecycle of Hurricane Katrina. The successful completion of this work is of great reference significance for future application development of Exa-scale computing systems, and provides valuable experience in programming, optimization methods and so on.
Innovation-driven development helps the great construction of Socialism with Chinese characteristics
On November 25th last year, during a panel discussion with members who won the 2016 Gordon Bell Prize based on the work titled “a fully implicit solver for atmospheric dynamics”, Qiu Yong, the president of Tsinghua University, pointed out that Sunway TaihuLight and its application research are important demonstration of Tsinghua University’s service to national innovation-driven development and an outstanding achievement in the development of interdisciplinary research.
In the future, the research team at Tsinghua University will continue to develop high-performance computing applications in multidisciplinary fields such as earth system simulation, bio-medicine, machine learning, aerospace, industrial manufacturing, new materials and new energy research etc., on Sunway TaihuLight. The supercomputer will play a more active role in helping build world-class universities, implementing the innovation-driven development strategy and accelerating the building of an innovative China.