Tsinghua Crowned Eight-Time Student Cluster Champions at ISC

This blog was originally published in HPCwire.

Always a hard-fought competition, the Student Cluster Competition awards were announced Wednesday, June 21, at the ISC High Performance Conference 2017. Amid whoops and hollers from the crowd, Thomas Sterling presented the award to the team with the best overall score, Tsinghua University. Sponsored by Inspur and using Nvidia graphics processors, Tsinghua was also one of the three teams who had a perfect score for the deep learning part of the competition.

The team is a force to be reckoned with. They are the only team to achieve a “triple crown” victory by winning competitions at ISC, SC, and ASC. They’ve won eight championships in total and this is their third win at the ISC competition.

Rounding out the list for the overall winners in this “very close competition” was 2nd place for Centre for High Performance Computing (CHPC) and 3rd place for Beihang University.

The award for Fan Favorite, the second year in a row, went to Universitat Politècnica De Catalunya Barcelona Tech (UPC) who captivated ISC attendees and garnered the most votes. Over 2,100 people voted for their Fan Favorite — a record for this conference attendee participation portion of the competition.

2nd place CHPC team with CHPC Director Dr. Happy Sithole and friends

The award for the Highest High Performance Linpack went to FAU boyzz from Friedrich-Alexander University Erlangen–Nürnberg. They are one of the few teams that has competed worldwide at all three competitions — the Supercomputing Conference (SC), Asia Student Supercomputer Challenge (ASC) in China, and ISC. They used a traditional cluster with 12 GPUs.

Onto the award for running deep learning applications, Vice General Manager XuJun Fu of Baidu Cloud, who brought the deep learning applications to the competition, announced the winners. Tsinghua University, Nanyang Technological University and Beihang University all took home the top prize for solving the Captcha Challenge, achieving the highest degree of model accuracy.

This year, ten teams from around the world came to Frankfurt to build a small cluster of their own design and test their HPC skills by optimizing and running a series of benchmarks and applications. The teams must keep their power consumption below 3000 watts on one power circuit while running the benchmarks and applications.

The teams used a variety of designs. Two teams utilized liquid cooling technology, eight teams used GPUs and one team used Xeon Phi. UPC built an ARM based cluster with 48 core chips, liquid cooled. EPCC University of Edinburgh (EPCC) were described as the Linpack junkies, driving their results with a liquid cooled system.

Gilad Shainer and Thomas Sterling reveal Tsinghua University as the team with the highest overall score

Scot Schultz, director of Educational Outreach, HPC Advisory Council, provided some additional perspective for this article. “We decided to use a slightly more interesting use case of solving for Captcha,” he said, “because it not only highlights the power of deep learning to be a useful tool create models to recognize and classify unwieldy data, such as distorted characters, grainy images and overlapping characters, but it also demonstrates that it is possible for this powerful technology to be used in less positive ways, such as solving for security or privacy.  Realizing that everyone has access to the tools we use to move society forward, we need to be aware of the possible mis-use, especially as it becomes more pervasive across industry, healthcare, financial services, and the like.”

“We’re always amazed and impressed by the quality of work and solutions we see from students during the ISC Student Cluster Competition,” said Doug Miles, director of PGI Compilers & Tools at Nvidia, a key sponsor for the student teams, many of whom rely on Nvidia GPUs to drive their workloads. “We support this event to encourage students from around the world to dive in and learn how to program parallel systems using the same compilers and programming models used by HPC professionals every day on the world’s fastest supercomputers.”

Gilad Shiner, chairman of HPC Advisory Council, said, “The HPC Advisory Council is proud to host, together with ISC, the 6th student cluster competition at the ISC High Performance conference. It enables the next generation of HPC professionals and drives HPC technologies to be used in more areas and applications.

“We want to thank all of the university teams that participated in the competition. Through this competition they have gained knowledge and expertise in HPC, deep learning technologies, and solutions. We hope to see these teams as well as new teams at the ISC 2018 competition.”

 

About the author: Kim McMahon is the CEO of McMahon Consulting, a full service marketing firm with over 15 years of experience in HPC, Enterprise Technical Computing, and the high-end IT space working with clients around the globe. Kim has performed sales and marketing for more years than she cares to count. She writes frequently on marketing, life, the world and how they sometimes all come together.