SCC11: Meet the Boston U Terriers
More cores, more better
Their four-node, dual-socket AMD Interlagos-based cluster has a grand total of 336 cores – 80 more than the next two teams. But they didn’t stop there. They also packed up four of NVIDIA’s latest M2090 Tesla GPU accelerators just to give them that extra lift when the number crunching gets thick.
Throw in some memory (32 GB per node, maybe on the light side), some QDR Infiniband, and maybe a few jimmies on top to add some crunch. At the end of the day, you have the biggest assembly of CPU and GPU power at the SCC this year.
In another first, they’re the only team to pit potential team members against each other to earn spots on the roster. Plus they have the first set of identical twins ever to participate in the SCC.
Most competition observers aren’t sure how this twin strategy will factor into the competition, but one of the Boston team members said that the twins operated in perfect parallel, exhibiting high bandwidth and very low latency.
We’ll see how their massive hardware and spooky, twin-based approach to parallelism plays out as the competition begins. The first test of this will be when LINPACK results are released Tuesday afternoon. Stay tuned … ®
Sponsored: Hyper-scale data management