On Friday, April 28, 2017, the winners of the sixth annual Asia Student Supercomputer Challenge (ASC17) were revealed against the backdrop of TaihuLight, the world’s fastest supercomputer, on the first floor of the National Supercomputing Center in Wuxi, China.
Over a four day span, 20 finalist teams competed to build their cluster and run seven applications: the High Performance Linpack benchmark, the High Performance Conjugate Gradient (HPCG) benchmark, a third-generation genomic assembly code Falcon, a traffic prediction problem with PaddlePaddle, the mystery application Saturne (an open source CFD code), numerical wave modeler MASNUM on the TiahuLight supercomputer, and the molecular dynamics code LAMMPS on a separate “Knights Landing” system. All applications run on the students’ clusters must stay within a strict 3,000 watt power envelope. Teams must also deliver a presentation that comprises 10 percent of their overall score.
And the winners are…
Tsinghua University is this year’s grand champion and will take home a cash prize of 100,000 CNY (~$14,500). The team also won the e-Prize (27,182 CNY, equivalent to $3,940), which required teams to complete a deep parallel optimization of the high-resolution maritime data simulation code MASNUM on TaihuLight. Tsinghua expanded the original program to 10,000 cores and sped up the program by 392 times, according to contest officials. MASNUM was one of three entries based on results from the Sunway TaihuLight machine nominated in 2016 for the Gordon Bell Prize. Tsinghua also received the top score for its optimization of Falcon.
Tsinghua University is not new to the student cluster competition circuit. In 2015, the team proved a triple threat: picking up all three contest titles: ASC15 (Student Supercomputer Challenge 2015), ISC15 (International Supercomputing Conference 2015) and SC15 (Student Cluster Competition 2015).
Beihang University scooped up second prize, scoring highest marks on both the general-purpose CFD code Saturne and the AI traffic optimization problem that used Baidu-created open source code PaddlePaddle. The team’s deep neural network model delivered the most accurate prediction of road conditions during the morning peak drive time. Its optimization of Code_Saturne yielded a 5X faster speedup over the average (of non-zero scorers). Beihang will receive a cash prize of 50,000 CNY (~$7,250).
Both Tsinghua and Beihang upgraded their Inspur-provided servers with eight Tesla P100 GPUs.
The winner of Highest Linpack (~$1,450 prize) was first-time finalist Weifang University, which achieved a new world record of 31.70 double-precision floating point teraflops on a heterogeneous five-node, ten-P100 GPU cluster. The previous record of 31.15 teraflops was set by the University of Science and Technology of China (USCT) team at SC16. Weifang also produced the highest score on HPCG benchmarking application: 992.333 Gflops, another student contest record.
Another new team University of Warsaw came in a close second on the Linpack testing (with 29.79 teraflops) using eight P100s in four nodes, and performed admirably across the application set. They will be a team to watch and we hope to see them at the upcoming ISC and SC events.
The four winning teams selected for the Application Innovation award (~$1,450 prize each) were Ural Federal University, National Tsing Hua University, Northwestern Polytechnical University and Shanghai Jiao Tong University. The two Best Popularity awards went to Saint Petersburg State University, whose fans took to Twitter in a show of support, and Zhengzhou University, which got lots of love on the Chinese-based WeChat platform. Each team will receive USD $500.
ASC17 in Photos
Please feel free to use or share any of these photos with attribution to HPCwire.