HPC Blog We’re only a few weeks away from the 6th annual ISC Student Cluster Competition, and student cluster aficionados worldwide are gearing up to watch the battle. ISC Cluster Competition Taipan Pak Lui and I shot a video at SC16 where we unveiled the teams for the ISC event, you can watch it below.
For those of you who don’t have the patience for the video (jerks), here’s the list of the teams along with a little colour commentary (there’s more colour and colourful commentary in the video).
Centre for High Performance Computing (CHPC South Africa): This will be the fifth time the hardy South Africans have competed at ISC. They have a fantastic record, winning three times and coming in second once. Can they go the distance and bring home the gold a fourth time? We’ll see.
Nanyang Technological University (Singapore): The last time we saw Nanyang at a big time cluster competition, they were competing in Wuxi at the ASC16 event. The year before that, they scored the highest LINPAC at ASC15. This is a team that’s on the brink of breaking through to the elite, is this the year they’ll do it?
University of Edinburgh (EPCC, UK): Now here are some LINPACK jockeys. This team set new LINPACK records in both the 2013 and 2014 ISC competitions. They were driving liquid cooled hardware in both cases. It’ll be interesting to see if they’re aiming at LINPACK again this year or if they’re setting their sights on the Overall Championship.
NUDT + USTC (China): This is a scary-smart team, a combination of China’s National University of Defense Technology and the University of Science & Technology of China. NUDT is a long time cluster competitor, winning three LINPACK awards along the way. USTC isn’t quite as experienced, but they just took home the Overall Championship at SC16. The combined team is one that should be feared by other competitors.
Freidrich-Alexander Universitat (FAU, Germany): This is the fourth year this German team will be competing at the highest levels of student clusterdom. But it’s the first time they’ll have a home court advantage. In the past, they’ve always traveled to the US for SC competitions, but this year they’re staying close to home. The team has had its ups and downs with finishes in the middle of the pack. Maybe the lack of jet lag will help them out this year.
University of Hamburg (Germany): Another home team for host country Germany, the Hamburgers have steadily improved since they started competing in 2014 at ISC. Their first cluster looked like it was housed in an accordion case, but they brought more hardware in year two, with an increase of performance. The team to date has eschewed the use of accelerators, which puts them at a disadvantage in some applications, but maybe they’ve gotten the GPU religion in the past year. They either have to bring more nodes or use accelerators in order to move into the elite tier of teams.
NERSC (US): This is the second time the National Energy Research Scientific Computing Center has fielded a team at ISC. Like the previous team, this one is composed entirely of female students – including two students who are high school students. This is also a far-flung team, with members from Puerto Rico, Boston, Texas, and Missouri.
Universitat Politencnica De Catalunya (Spain): It’s the third time around for the Universita…Politenci…Team Spain. In the past two competitions, they’ve brought high node count clusters populated by many ARM chips, huge number of cores, and some accelerators. I always thought this team could be a contender if they had some halfway decent GPUs behind them. In their last outing, they had 6 NVIDIA K80s, I think they need more this time around….like maybe 8-10 P100s.
Purdue University + Northeastern University (US): This is another combined team, this time with the corn fed midwestern Perdue kids teaming up with the city-wise Northeasterners from Boston. Can these two different teams combine to make cluster competition magic? They’re both experienced programs, with Purdue having competed in many event and NEU having a hand in at least three. But both programs have typically finished in the middle of the pack. Maybe combining their brains will power them into the upper echelon this year.
Boston Green Team (US): Yep, it’s another version of Team Chowder, fresh from Boston. This time it’s a team culled from Dartmouth University and UMass Boston. These teams are always thinking out of the box, which leads to some unusual boxes. They’re the first team to try fat nodes, with their Fatal Four Way system a previous competition. They were also the first team to use Jetsons in their cluster. It’ll be interesting to see what they’re sporting this year.
Beihang University (China): This is the second place team from the ASC17 competition held earlier this year. The first two Chinese finishers at ASC get an automatic bye into the ISC competition. Beihang turned in a fairly strong performance at ASC – strong enough to beat out 18 other teams for their golden ticket to Frankfurt. The team hasn’t competed internationally yet, so they’re going to have some jet lag and cultural differences to deal with. But there’s nothing like a hugely long airplane trip across many time zones to focus the mind – or drive you out of it.
Tsinghua University (China): This is the second ASC17 team and the one who won the Overall ASC Championship earlier this spring. They blew away the rest of the field with a dominant performance vs. the other 19 teams. They were solid on all of the applications, but were bested on the LINPACK and HPCG benchmarks. Otherwise, this Tsinghua team is firing on all cylinders. They’re also highly experienced, having participated in many past competitions (and winning several). In fact, Tsinghua was the only team to win all three major competitions in a single year. Along with CHPC, Tsinghua has to be a favorite at ISC17.
As always, we’ll be covering the ISC17 Student Cluster Competition like fur on a weasel. I’ll be there to bring all the action into your offices and living rooms. Next up we’ll talk about the applications and challenges facing this crop of student clusterers.