Students schlep from afar to join the quest for international HPC glory
El Reg profiles this year's overseas cluster confab entrants
SCC'13 Three teams of students will be jetting in from overseas to enter this year's Student Cluster Competition, due to kick off on November 18 at the annual SC13 Supercomputing Conference.</p?
The teams will be competing in the Big Iron division where undergraduate students design, build, and tune their very own HPC clusters. They’ll be running HPC benchmarks and scientific applications to see which team has the fastest system and has done the best job of learning about scientific computing.
There are a total of eight teams vying for the Big Iron Overall Championship crown and the Highest LINPACK Sceptre of Honour (there isn’t an actual sceptre, unfortunately.)
Let’s take a look at the teams who are travelling the farthest:
Team Australia: Australia’s iVEC, a joint venture between the country’s national science agency and several of its universities, is fielding the first team from Down Under. They’re not only Down Under, but very Far Away too, given their location in Perth – 8,700 air miles from Denver, the SC13 venue.
The idea for forming Team Australia (which perhaps should be dubbed Team Dingo as in, “The dingo ran your LINPACK”) came from iVEC representatives attending SC12 in Salt Lake City who observed that cluster competition. They see the contest as a way to foster home-grown HPC talent from Australian universities.
Team Dingo has gone through “the ultimate training schedule,” in the words of their application to enter the competition. In addition to taking academic courses, the Aussie students have received special training from iVEC personnel, as well as catching the eye of hardware sponsor SGI’s top HPC expert in the Asia Pacific region.
The Dingoes certainly bring a lot of enthusiasm into the competition. They’ve been updating their activities and progress on Twitter. They got our attention by tweeting about their “seriously awesome kit from SGI” and the arrival of their Nvidia Tesla GPUs back in August.
Team Germany: The other rookie team from afar hails from Germany’s Friedrich-Alexander University of Erlangen-Nuremberg (FAU). FAU is one of Germany’s top research universities, ranked as the 179th best Engineering & Technology School on the 2013 QS university ranking scale.
The FAU application was one of the better ones that I’ve seen over the years, leading the reader through the criteria they used to evaluate their hardware choices. They even cited Amdahl’s Law (PDF, 4 pages), which is used to find the maximum expected improvement to a system when a component is improved or upgraded.
The German students very clearly outlined the arguments for and against thin node (single CPU) vs. fat (quad-socket) nodes, touching on inter-node communication needs and the cost in terms of energy usage.
They also talked about accelerators in detail, including their decision criteria and a discussion of the trade-offs involved in deciding the balance of CPUs to accelerators per node.
I didn’t see any information on the composition of the team, their individual experience, or how they are approaching the application tasks. However, I did discover that German HPC vendor Megware is sponsoring the team, which means they are in pretty good hands.
Megware was the sponsor behind the Chemnitz University “Coffee Table of Doom” that competed at the ISC13 competition in Leipzig last summer. Given this, I would expect to see Team Germany arrive in Denver prepped to the hilt – or even higher.
Team China: To avid followers of SCC competitions, China’s National University of Defense Technology needs no introduction. NUDT finished second overall in their first bout at SC11 in Seattle. They placed second again at ISC’12 but topped the field in LINPACK. At SC12 in Salt Lake City, they placed – yes, second again – and turned in the highest LINPACK score of 3.014 teraflops/s.
They were one of the first teams to effectively utilize GPUs, optimizing non-GPU-centric apps for GPU acceleration. This was quite a surprise to observers in 2011, but given the heritage of the team (they get to practice on Tianhe-1, which was 2010’s #1 on the Top 500 supercomputers list), it isn’t hard to understand why this is a team that should be seen as a serious competitor.
Although I haven’t seen the NUDT entrance application, I’m sure they’re bringing some competition veterans to Denver. I’m expecting that they will bring a Xeon system backed by a heaping helping of NVIDIA Tesla accelerators. I would also guess that long-term sponsor Inspur is once again supporting NUDT’s efforts.
In our next article, we’ll take a look at the US home teams and see how they stack up against the international competition.