Props to group top rated flops.
Digital this calendar year, the SC20 University student Cluster Levels of competition was continue to all about groups vying for top supercomputing efficiency in the yearly battle for HPC bragging rights.
That honor went to Beijing’s Tsinghua University, whose 6-member undergraduate pupil group clocked in 300 teraflops of processing efficiency.
A a single teraflop personal computer can course of action a single trillion floating-position operations for each 2nd.
The Digital College student Cluster Opposition was this year’s battleground for 19 teams. Opponents consisted of either substantial faculty or undergraduate pupils. Groups were built up of six members, an adviser and vendor associates.
In the 72-hour levels of competition, student groups created and constructed digital clusters jogging NVIDIA GPUs in the Microsoft Azure cloud. Learners accomplished a set of benchmarks and true-environment scientific workloads.
Teams ran the Gromac molecular dynamics application, tackling COVID-19 investigate. They also ran the CESM application to perform on optimizing local weather modeling code. The “reproducibility challenge” identified as on the teams to replicate final results from an SC19 analysis paper.
Among other hurdles, groups had been tossed a surprise exascale computing task mini-software, miniVite, to exam their chops at compiling, managing and optimizing.
A leaderboard tracked functionality success of their submissions and the amount of money used on Microsoft Azure as well as the burn up level of their paying out by the hour on cloud means.
Roller-Coaster Computing Problems
The Ga Institute of Technological innovation competed for its next time. This year’s squad, dubbed Group Phoenix, experienced the superior fortune of landing advisor Vijay Thakkar, a Gordon Bell Prize nominee this 12 months.
50 percent of the group members ended up teaching assistants for introductory programs classes at Georgia Tech, said crew member Sudhanshu Agarwal.
Ga Tech employed NVIDIA GPUs “wherever it was achievable, as GPUs decreased computation time,” mentioned Agarwal.
“We had a good deal of enjoyable this 12 months and search forward to taking part in SC21 and further than,” he said.
Pan Yueyang, a junior in laptop or computer science at Peking University, joined his university’s supercomputing crew just before getting the leap to take part in the SC20 struggle. But it was entire of surprises, he noted.
He claimed that all through the competitors his staff ran into a collection of unexpected hiccups. “Luckily it concluded as expected and the funds was a bit underneath the limitation,” he explained.
Jacob Xiaochen Li, a junior in computer system science at the College of California, San Diego, mentioned his crew was relying on NVIDIA GPUs for the MemXCT portion of the opposition to reproduce the scaling experiment along with memory bandwidth utilization. “Our final results match the original chart intently,” he mentioned, noting there had been some hurdles alongside the way.
Po Hao Chen, a sophmore in computer system science at Boston University, said he fully commited to the level of competition mainly because he’s often loved algorithmic optimization. Like a lot of, he had to juggle the level of competition with the demands of courses and tests.
“I stayed up for three full days doing work on the cluster,” he said. “And I seriously discovered a good deal from this opposition.”
Teams and Flops
Tsinghua University, China
Southern College of Science and Technological innovation
Texas A&M University
Ga Institute of Technology
Nanyang Technological University, Singapore
College of Warsaw
University of Illinois
Massachusetts Institute of Engineering
College of California, San Diego
North Carolina Point out University
Friedrich-Alexander College Erlangen-Nuremberg
Shanghai Jiao Tong University
University of Texas
Wake Forest College