“The Student Cluster Competition is a terrific event, where the students’ object is not winning, but doing well, learning from the experience, and using what they learn in their own education,” said Art Mann of Silicon Mechanics.
Bothell, WA (PRWEB) November 07, 2012
Silicon Mechanics, Inc., a leading manufacturer of rackmount servers, storage, and high-performance computing solutions, announces that it is sponsoring the Massachusetts Green High Performance Computing Consortium (MGHPCC) in the Student Cluster Competition at the SC12 Conference & Exhibition, taking place November 10-16, 2012 in Salt Lake City, Utah. The team, composed of undergraduates from Boston University, Harvard University, MIT, University of Massachusetts, and Northeastern University, will complete in a real-time, non-stop 48-hour challenge to demonstrate the best sustained performance on a series of applications.
The Student Cluster Competition showcases student expertise in deploying and using a high-performance cluster in four pre-determined applications. Students are given the application areas in advance to familiarize themselves and test their codes, but will not know the actual task until the starting whistle is blown. The MGHPCC team has been busy practicing, optimizing the contest applications, independent of configuration.
This year’s 6-member team is comprised of computer engineering, electrical engineering, computer science, and physics majors. Two additional students will observe, learning the ropes and preparing to take over for graduating seniors in next year’s contest.
Silicon Mechanics has configured an HPC cluster, valued at $80,000, comprised of hardware provided by partners NVIDIA, AMD, Mellanox, Supermicro, Kingston Technology, and Seagate. Equipped with NVIDIA® Tesla® GPUs and AMD Opteron™ 6300 Series processors (code named Abu Dhabi), the HPC cluster contains a head node, eight compute nodes, two GPU nodes, and Gigabit and InfiniBand networking.
As the MGHPCC team’s hardware sponsor, Silicon Mechanics was responsible for identifying hardware partners and securing their commitment, assembling and testing the cluster, and then shipping it to the consortium for use in the competition. “The student team is restricted to a power budget of 26 amps, and had an opportunity to collaborate with Silicon Mechanics on tuning parameters and discussing options that would get the most out of the allowed power budget,” said the group’s advisor, Daniel Kamalic, Director of Systems Analysis and Administration in the Department of Biomedical Engineering at Boston University.
This year’s competition application areas include classical molecular dynamics, using LAMMPS (Large-scale Atomic/Molecular Massively Parallel Simulator); quantum dynamics, using QMCPACK, a framework for quantum Monte Carlo simulations; climate modeling, using the Community Atmosphere Model (CAM); and reactive flow modeling, with PFLOTRAN, an open source massively parallel subsurface flow and reactive transport code.
The student team had no prior knowledge of any of these applications and found it to be an exciting challenge, given the real-world relevance of the applications. They even took some time to re-code the applications to optimize them so they’d run faster. Team member John-Nicholas Furst notes, “The experience is greatly additive to the things we may learn in class, where we concentrate on theory and good practice. This brings it home to the level of the type of real world situations where these expensive, powerful high-performance machines are used. It’s exactly what we can expect when we go into industry after college.”
The teams will run each task several times, tweaking their code to reduce their times as much as possible, and then submit their results to the judges. The competition has an Olympics time-trial atmosphere, with winning times varying by mere fractions of a second. Five of the six team members competed last year, where they performed well. They are banking on their prior knowledge of the competition and yearlong preparation to reduce the anxiety factor and improve their performance even further. “Winning is a nice side effect, but the collegiality, camaraderie, and real world challenge is what it’s really all about,” said Furst. “We will try our hardest to do the best we can, while also talking to the other teams, discussing best practices, and offering hints of things to look out for.”
The winning team will be chosen based on a combined score for workload completed, benchmark performance, interviews and outreach. Recognition will also be given for the highest LINPACK and SCC Fan Favorite. The awards for the Student Cluster Competition will be announced at the SC Awards Ceremony on Thursday, November 15, 2012.
“The Student Cluster Competition is a terrific event, where the students’ object is not winning, but doing well, learning from the experience, and using what they learn in their own education,” said Art Mann, Silicon Mechanics’ education/research/government vertical group manager. “Another unique factor is that teams usually include students with a variety of expertise, often involving students majoring in the sciences, engineering, computer sciences, or business.”
About Silicon Mechanics
Silicon Mechanics, Inc. is a leading manufacturer of rackmount servers, storage, and high-performance computing clusters, with one of the industry’s most comprehensive product offerings. Working collaboratively with customers to create best-fit solutions at competitive prices, we support our products with superior warranty offerings and a team of sales and service engineers dedicated to service and solutions excellence. Based in Bothell, WA, Silicon Mechanics is an Intel® Premier Provider and a Premier member of the AMD Fusion Partner Program.
Additional information about Silicon Mechanics may be found online at http://www.siliconmechanics.com, or by contacting info(at)siliconmechanics(dot)com.