When engineering professor Dave Kaeli approached Neel Shah, E’15, and Tushar Swamy, E’15, about entering a supercomputer competition, the students’ combined experience in the task at hand was zero.
“I had never even built a regular computer,” says Shah, let alone a super one. But that didn’t stop him from signing up with Swamy—and from stealing the show.
In collaboration with three students from Bentley University, the Northeastern duo built a supercomputer within the financial and power constraints outlined by the Student Cluster Competition, an event that brings together teams of university students for a live face-off to see who has built the fastest computer.
Seven other teams, hailing from as far away as Australia, also competed at the two-day event in Denver.
One key to Northeastern’s success was choosing unconventional hardware. Instead of building their computer with a central processing unit alone, they also used a graphics processing unit. Both Shah and Swamy had four years of undergraduate research experience in Kaeli’s lab, where they learned the ins and outs of the GPU. Although this specialized processor is used to handle images in Kaeli’s lab, it’s also used to process massive data sets. The approach gave the team solid power efficiency while also bringing down the cost.
In keeping with a shift toward efficient high-performance computing, this year’s competition had two tracks: one restricted only by power, and one restricted by both power and cost. Shah and Swamy’s team entered the latter.
To achieve high performance without spending a lot, they used a low-cost advanced platform called an APU, or acceleration processing unit. Made by Advanced Micro Devices, the APU combines both CPUs and GPUs on one board. The approach won the team first place in the combined class. It also stood up well against the teams that had no spending constraints. When competing against computers that cost in the six figures, say Shah and Swamy, the Northeastern/Bentley computer, which cost less than $2,500 to produce, came in fourth place.
The students credit their victory to a combination of strategy and technical innovation. Although they knew ahead of time which applications their computer would need to run during the competition, they also had to prepare for a wild card—a mystery program that wouldn’t be revealed until they set foot in the conference hall. They also made sure to keep a little of the computer’s processing power available to use for queuing up the next program in the competition.
It was their combination of good time management, cross-institutional collaboration, and thinking outside the hardware box that brought them their success, says Swamy.
Also new on the blog …
Angela Herring is Northeastern’s science writer. Contact her via twitter (@NU_iNSolution), at her blog, iNSolution (northeastern.edu/insolution), or at email@example.com.