On April 23, the third day of the ASC19 Student Supercomputer Challenge (ASC19) final held at Dalian University of Technology, the top 20 teams accomplished HPL/HPCG, Community Earth System Model (CESM) and the first-ever Group Competition. In the HPL benchmark, Beijing University of Aeronautics and Astronautics (BUAA) set a new record of 50.21 TFLOPS within 3000W power consumption, up by 17% compared with last year’s 42.99 TFLOPS. Sun Yat-sen University (SYSU) leads in the CESM challenge. And the group made up of teams from SYSU, Taiyuan University of Technology, the Chinese University of Hong Kong and Friedrich-Alexander-Universität Erlangen-Nürnberg become the winner of the first-ever Group Competition.

BUAA Set a New HPL Record

As the most common supercomputer benchmark in the world, HPL is an important indicator for the TOP500 List. The HPL test of the ASC competition should be finished within 3000W power consumption.

The BUAA team designed a heterogeneous supercomputing system with 3 servers and 12 cards – 3 of Inspur NF5280M5 AI supercomputing servers supporting NVIDIA Tesla V100 GPU accelerators, with which their consistent floating-point performance reached 50.21 TFLOPS after optimization. Jinan University, a “black horse” entering the ASC finals for the first time, ranked the second with 39.68 TFLOPS. Taiwan Tsing Hua University, Tsinghua University and Dalian University of Technology all surpassed 30 TFLOPS.

The HPL benchmark is a big challenge due to its demands for the highest possible computing performance under power-limited conditions. Yang Hailong, the advisor who had led the BUAA team in the ASC competition for six times, had a deep understanding in this regard. He believed that the ASC competition is full of uncertainty even with good preparation. “All the participants are undergraduate students. Very few of them get to know supercomputers, let alone designing a solution and building a supercomputing system on site. In this process, they will face many problems that they have never encountered. Therefore, how to cope with and solve unexpected problems is important to every team member,” he said. “Participating in the ASC competition is actually a process of combining theory with practice and converting knowledge to capacity, which is of high value for students’ future career development.”

Taiwan Tsinghua University ranked first in the HPCG benchmark with 2004.96 GFlops, followed by BUAA and Beijing University with 1529.20 GFlops and 1483.70 GFlops respectively. HPCG, or high-performance conjugate gradient benchmark, has now become a supplement to HPL benchmark, used in the TOP500 List. Compared with HPL, HPCG is more in line with supercomputing applications, aiming to achieve computing and data access models matching with different HPC applications, and motivate designers of supercomputing systems to continuously improve the performance of applications.

SYSU Leads in the CESM Challenge

CESM, Community Earth System Model, is the most widely used climate model in the world today and one of the main climate models adopted by the United Nations Intergovernmental Panel on Climate Change (IPCC) for the fifth and sixth assessment reports. CESM is a coupled climate model including major modules for atmosphere, land, ocean, sea ice, and land ice. Based on observed climactic information as an initial state, it uses the physical, chemical, and fluid dynamic and other equations to replicate the process of climate change on supercomputers. The winner of the CESM Challenge will get the e Prize.

CESM task is a huge challenge for computing performance. It requires all teams to finish the computing of two atmospheric models, CAM4 and CAM5, and the calculation of the whole model under the condition of double carbon dioxide concentration. Only by understanding and optimizing CESM can the teams shorten calculation time while ensure accuracy. In the end, Sun Yat-sen University achieved the best performance.

Zhou Tianjun, representative of supporting organizations of the CESM challenge, researcher and assistant to the director of the Institute of Atmospheric Physics of the Chinese Academy of Sciences, and deputy director of LASG State Key Laboratory, noted that on one hand, the CESM challenge can make college students know about the latest international climate change project. On the other hand, introducing carbon dioxide concentration change into the final will make everyone feel global warming caused by increasing carbon emissions and its impact on the environment, so as to raise our environmental awareness.

Multinational Teams Won the Group Competition

The group made up of teams from SYSU, Taiyuan University of Technology, the Chinese University of Hong Kong and Friedrich-Alexander-Universität Erlangen-Nürnberg become the winner of the first-ever Group Competition.

For the first time, the ASC19 final contains a Group Competition. The top 20 teams formed 5 groups through an on-site draw and tackle the Fluidity challenge through cross-team cooperation. Fluidity is a three-dimensional software package for computational fluid dynamics, widely used in many fields, such as geophysical fluid dynamics, computational fluid dynamics, ocean modeling, and mantle convection. It is based on the finite element/control volume method which allows arbitrary movement of the mesh and changes in mesh resolution with time dependent problems. It is parallelized using MPI and is capable of scaling to thousands of processors.

“Cross-team cooperation is very common in research, especially in international science and engineering projects. Our intention of setting Group Competition is to encourage college students to join hands and complete the task.” said Liu Jun, a member of ASC19 organizing committee, “Why Fluidity was chosen is that currently fluid mechanics is important for the development of high-end manufacturing, such as in aerospace, automobile, shipbuilding and marine engineering, etc. We hope the students’ innovative ideas can bring new enlightenment to HPC application in manufacturing.”

On April 24, all teams will face more challenges. The artificial intelligence (AI) task, Face Super Resolution is a domain-specific super-resolution (SR) problem and a popular AI visual computing technology that allows computers to generate HR facial images that approximate authentic ones. The tasks on April 24 also include the third-generation gene sequencing assembly software WTDBG developed by Chinese scientists and a mystery application task that will be announced on site.

The ASC Student Supercomputer Challenge is the world’s largest supercomputer competition. ASC19 was jointly organized by Asia Supercomputer Community, Inspur Group and Dalian University of Technology. More than 300 teams from all over the world signed up for the competition and 20 of them entered the finals. The champion, Silver Prize, Highest LINPACK and e Prize of the finals will be unveiled on April 25.