The odds were stacked against the Center for High Performance Computing (CHPC) team from South Africa.
By all accounts, they had little chance of being competitive in, let alone winning, the 2013 Student Cluster Challenge held last week at the International Supercomputing Conference (ISC) in Leipzig, Germany.
Representing Africa for the first time in an international computing challenge, they lacked competition-tested experience. They also were the youngest of the eight teams, made up only of undergrads.
And worse, they couldn’t access their cluster until a day after they arrived at ISC, while the other seven teams had been practicing on their systems well ahead of time.
But once the dust settled and scores were tallied, CHPC stood alone at the top, earning the coveted “Overall Championship Award” in the competition.
Their secret sauce: NVIDIA GPU accelerators.
The Edge Goes to GPU Accelerators
Dan Olds, principal analyst with Gabriel Consulting and creator of the new Student Cluster Competition website, said that NVIDIA GPU accelerators made quite a difference:
“Like all the other teams, South Africa used accelerators to give their cluster a turbocharged punch. While some teams used Intel Phi co-processors, team South Africa went with eight NVIDIA K20 cards, which seemed to do the trick.”
Using an eight-node Dell PowerEdge system with eight NVIDIA Tesla K20 GPUs – the highest performance, most efficient accelerators ever built – Team South Africa achieved a runaway victory for the overall competition.
In the competition, teams were scored on how well their cluster performed on a number of real-world applications including GROMACS, a molecular dynamics package, MILC, a quantum chromodynamics app, and WRF, a widely used weather research and forecasting application.
There were also two ‘mystery’ applications that were disclosed on the day of the competition: the AMG numerical analysis application and the CP2K molecular simulation package.
Together, these performance marks accounted for 60 percent of South Africa’s overall score, with the rest coming from interviews with event judges.Team South Africa entered the competition as big underdogs, and emerged as the big victors.But they weren’t the only ones to benefit from GPUs.
All Accelerators are not Created Equal
Olds, in a recent blog post, discussed the dramatic increase in the use of accelerators by the winning student cluster teams:
“Everyone is using accelerators this year, either NVIDIA GPUs or Intel Phi co-processors. We’ve seen accelerator use increase steadily since they made their first appearance in 2010.”
In addition to South Africa, GPUs powered the top four systems earning the highest Linpack scores, soundly beating off other teams that used Xeon CPUs or Xeon Phi accelerators:
- Team Huazhong, China — 8.455 TF
- Team Edinburgh, UK — 8.321 TF
- Team Tsinghua, China — 8.132 TF
- Team South Africa — 6.371 TF
GPU accelerators gave these and other teams a competitive edge by delivering the highest performance per watt, which is critical given the competition’s 3,000 watt power cap.Fifth-Straight Win for GPUsThis is the fifth straight time that GPU-accelerated systems have won the “Overall Award” in the Challenge, as well as achieved the top Linpack scores. In addition to ISC 13, GPU-equipped teams came out on top at ISC 2012 and Supercomputing Conference 2011 and 2012, and the top three Linpack-winning systems in the first-annual Asia Student Cluster Challenge April 2013 used Tesla GPUs as well.Next up: the student competition at the SC13 conference in Denver this November. Will teams using GPU accelerators reign supreme again? Place your bets, and let us know what you think.