The preliminary competitions for the ASC19 Student Supercomputer Challenge have begun. 300+ teams from over 200 universities around the world will compete for the final Top 20 spots. They will tackle challenges in AI-application single-image super-resolution (SR), Community Earth System Model (CESM), and High performance benchmarks HPL and HPCG. To help the participating students become familiar with the competition, the organizing committee held a training camp in Beijing from January 21st to 22nd with numerous teams attending.
At the camp, HPC and AI experts from the CAS Institute of Atmospheric Physics, the State Key Laboratory of High-end Server & Storage Technology, Inspur, Intel, NVIDIA, and Mellanox explained the competition rules, and led training sessions on supercomputing skills. These included: cluster construction and evaluation, network optimization and selection, OpenACC programming and MPI programming, exploring SR, CESM tasks and how to solve them. Other lectures taught AI utilization in deep learning and neural networks. The training camp also invited the ASC18 champion team from Tsinghua University, and the highly ranked team from Huazhong Agricultural University to share their experience in the competition. They provided insight into the construction and optimization of heterogeneous environments, understanding and optimization of applications in multiple domains, screening and optimization of serial and parallel instructions.
In addition to providing free training for participating teams, the camp also allows team members to observe super-computing clusters in real-world applications. Many participants expressed appreciation for the opportunity to deepen their comprehensive understanding of supercomputers.
The ASC Student Supercomputer Challenge is the world’s largest supercomputer competition for undergraduates. Known as the most challenging event in its field, the difficulty of the ASC will be raised once again this year. The SR task requires participants to design and optimize algorithms based on the PyTorch framework, and perform 4x SR upscaling for 80 images which were down-sampled with a bicubic kernel, while improving the quality of the reconstructed high-resolution images as much as possible. The CESM test requires the participants to optimize the Community Earth System Model (CESM), which has numerous lines of code, to reduce the computational needs in time and resources while simulating climate changes over a decade based on historical data.
Participants in this year’s challenge include previous ASC champion teams such as Tsinghua University, National Tsing Hua University, Nanyang Technological University, as well as new-comers like Sungkyunkwan University and Monash University. The numerous talented teams promise an exciting competition for the Top 20. The finals will be held at Dalian University of Technology from April 21-25.