A 48-hour supercomputing battle pushes students to question decisions. GPUs aren’t one of them.
At the SC18 supercomputing show’s Student Cluster Competition, Tsinghua University, of China — running eight NVIDIA V100 GPUs — snagged the crown for overall winner.
The honor of highest Linpack score — a measurement of a system’s floating point computing horsepower — went to Nanyang Technological University, in Singapore, which achieved 56.51 teraflops.
“If you don’t have GPUs, best of luck. It’s essential,” said Bu-sung Lee, team leader and faculty adviser at Nanyang, which won the competition’s top honor last year.
The 16 teams of six members from around the globe came to compete in the world’s premier supercomputing event, held in Dallas this year.
The competition — which limits students to using 3,000 watts of power — was grueling. Teams hunkered down during battle, staring intensely into laptops. Take-out food boxes delivered by their coaches, who weren’t permitted to interact with the students, littered their booths.
In an endurance race that goes around the clock for two days straight, students tackle real-life computing workloads. “They are taking turns staying up through the night, three at a time,” said Lee.
The Student Cluster Competition gives students interested in high performance computing a chance to deploy their workloads on supercomputing systems and mix with others in the industry, encouraging the pursuit of careers in the field.
Students have duked it out in teams over these supercomputer cluster competitions since the Supercomputing Conference of 2007. That first SC07 battle, held in Reno, Nevada, has inspired student competitions in Germany, China and South Africa.
The clusters are designed to take on specific tasks already faced by those deploying high performance computing.
All teams were offered the chance to run their clusters on our NVIDIA V100 32GB GPUs, which we donate to contestants each year. “It’s rare to be able to play with the latest tech,” said Lee.