For the sixth consecutive time, a system powered by GPUs captured the top-spot overall in the Supercomputing conference’s Student Cluster competition.
With longhorn hand-signs and smiles flashing, an 11-person team from the University of Texas, at Austin, claimed the award before a crowd of more than 500 at the end of three-day show in Denver.
It’s no accident that a GPU-based system won the competition. Seven of the eight teams from around the world that finished the grueling 48-hour, nonstop competition utilized Tesla K20x accelerators donated by NVIDIA.
That’s a powerful testament to the grip that parallel processing has on the next generation of computer scientists.
Julian Michael, a junior on the team, said he and his undergrad classmates – all computer science majors with a computational biologist thrown in — had been preparing since March for the competition, putting in 10 hours a week and sleeping just four hours a night toward the end.
“We’re going to go back, get some sleep and get ready for next year’s competition,” he said.
There’s a definite minimalist slant to the competition. Contest rules require that the teams build a supercomputing cluster working with donated equipment and an austere power budget of 3,000 watts, enough to power a pair of microwave ovens.
Scoring takes into account how each system performs on a range of standard HPC applications – such as WRF, used for weather prediction, and GraphLab, a tool for big-data analysis – as well as on extensive interviews.