In this article, we take a quick look at a few teams who out-performed expectations at the recently concluded ASC19. These are teams who haven’t been in the winner’s circle lately (or ever) but who deserve recognition for what they were able to do at the competition in Dalian.
First up, we have to give it to Team Beihang for winning the Linpack award. While their score of 50.21 wasn’t a worldwide Student Cluster Competition record, it was a record result for the ASC competition, and that’s saying something.
I briefly interview the Beihang team leader in the video below. I probe and press him on the topic of whether the team was just going for the Linpack prize or if they’re going for the Overall Championship as well. He steadfastly states that they’re going for all the marbles and are not just a Linpack stunt team. Given his explanations, and their results on the rest of the applications, I believe him.
We also need to recognize Shanghai Jaio Tong University for finishing second on the cursed CESM application. This climate modeling app was the bane of most teams and one of the most difficult applications in any cluster competition.
I was going to interview the winner of this application, Team Sun Yat-Sen, but when I told them they won, they all scurried off to see their scores on the big board and left me all alone holding my microphone. Fortunately, the kids from Shanghai were around and I managed to get an interview with them.
I also want to give a shout out to Peking University for taking down third place on HPCG. While HPCG wasn’t the most challenging application to get running, it does require some skill to optimize it for the various system configurations you see in a cluster competition.
It’s not all about the hardware and particularly not all about the GPUs. Team Peking, with eight nodes and 12 GPUs, should have finished behind several teams who had the same number of GPUs (or more) and fewer nodes.
In the video, we talk with Team Peking’s student leader and get his perspective on their performance. He sort of shrugs off the fact that there is skill involved in tuning HPCG – I can’t tell if he’s just being modest or if he truly believes that it’s just a matter of mathematics. You be the judge…
I can’t finish this article without throwing some love towards Team EAFIT. This is a team that saw both the lowest lows and their teams’ highest highs in just this one competition. Their low point was with LINPACK – their hardware turned into a steaming pile of, well, you know, during the period in which they had to run HPL, and they had to turn in a completely, well, you know, result.
Most teams would have been emotionally blown up by that kind of result. It would have put them into a mental hole they couldn’t crawl out of. But not Team EAFIT. The plucky Colombians firmly put HPL into their rearview mirror and kept pushing for the rest of the competition.
The end result is that the team placed sixth in the WTDBG2 application competition and scored an Application Innovation Award. This is the first award for an EAFIT team in six major cluster competition outings. For coming from a country with little or no HPC infrastructure, this is a pretty big deal. Good job, EAFIT.
Next up is a comprehensive look at who won at ASC19 and how they did it. This story will go into detail, and possibly feature the worst and most insipid sports clichés ever, so buckle up, it’s going to be quite a ride. Stay tuned…..