Huge cheers broke out in the New Orleans Theater inside the Ernest N. Morial Convention Center today when Team Longhorn from the University of Texas at Austin was declared the overall winner of the SC Student Cluster Competition for the third year in a row. Making the remarkable string of victories even more impressive, it was the first unaccelerated cluster win since 2011.
For the supercomputing community, student cluster contests are akin to college football championships, and this event was the biggest yet with twelve teams competing from five continents, listed below:
- The University of Texas – Austin
- Illinois Institute of Technology
- iVEC – Australia
- Friedrich-Alexander-Universität – Germany
- National Tsing Hua University – Taiwan
- University of Tennessee – Knoxville
- Purdue University /EAFIT – Colombia
- The University of Oklahoma
- The University of Science and Technology – China
- National University of Singapore
- Huazhong University of Science and Technology – China
- Massachusetts Green Team
The grueling two-day face-off takes place in real-time while teams of undergraduate and/or high school students build a cluster on the SC exhibit floor and race to demonstrate the highest sustained performance across a series of scientific workloads without going over a set power limit.
The winning teams for this year’s contest were revealed during the SC14 Awards Ceremony on Thursday, Nov. 20, where Team Longhorn proved its mettle one more time, winning in the overall category, and Taiwan’s National Tsing Hua University (NTHU) taking home the award for Highest LINPACK.
Under the Standard Track, six students partner with vendors to design and build a cluster from commercially available components without exceeding the 3120-watt power limit (26-amp at 120-volt). Two circuits, each with a soft limit of 13-amps, are provided to the teams. Teams can select any operating system and software stack that will run the challenges and display software. Domain experts may offer advice on tuning and running the competition codes.
The standard track was so termed to distinguish it from the commodity track, a new addition at SC13. Despite high interest in the track, which encouraged “off-the-shelf” and “off-the-wall” solutions as long as they were under $2,500 and stayed within a 15-amp power limit, it has been put on hold due to lack of funding. This underscores how important sponsorships are, and this year’s supporters include Bank of America, Chevron, Data Direct Networks, Geist Global, General Motors, MathWorks, Procter & Gamble, Schlumberger.
As with other years, the students must assemble, test, and tune their clusters as well as run the HPCC benchmarks before the opening bell. Directly after the ribbon-cutting ceremony on Monday night, signally the opening of the show floor, the participants are given the competition data sets for four applications, a set of real-world scientific workloads, three that they know about in advance (ADCIRC, NAMD, and MATLAB) and one that is a “mystery application,” which this year was Enzo, an advanced computational astrophysics code.
The primary objective of the competition is completing the greatest number of application runs during the 48-hour contest. Points are awarded for system performance, including the quantity and quality of results. Teams are also interviewed to test how well they understand the scientific workloads they’ve been running. The team with the most points is declared the “Overall Winner,” with a second award recognizing highest LINPACK score.
While not every team can win the contest, all of these students deserve our highest kudos for their skills and dedication. They dedicate hundreds of hours and travel long distances to get to the show, and they tend to get little sleep during the exhibit.
Be sure to check back with HPCwire in the coming weeks as we reveal more about the specifics of the winning machines and the runner-up teams.