729 teraflops, 71,000-core Super cost just US$5,500 to build
Cloud doubters, this isn't going to be your best day
Cycle Computing has helped hard drive giant Western Digital shove a month's worth of simulations into eight hours on Amazon cores.
The simulation workload was non-trivial: to check out new hard drive head designs, the company runs a million simulations, each of which involved a sweep of 22 head design parameters on three types of media.
In that context, HGST's in-house computing became a serious bottleneck, with each simulation run taking as much as 30 days to complete.
Hence, in what it describes as the largest enterprise cloud run so far, Cycle Computing spun up nearly 71,000 AWS cores for an eight-hour run.
Cycle Computing claims the cluster delivered 729 teraflops to run HGST's MRM/MatLab app under the control of CycleCloud cluster-creation software and Chef automation system.
The cloud outfit says it spun the app up from zero to 50,000 cores in 23 minutes, and calculates that the run, dubbed “Gojira”, completed nearly 620,000 compute-hours.
Cycle Computing also pointed at the advance in silicon in the last 12 months. The Gojira run, it says, ran 50 per cent better FLOPS per Ivy Bridge core than last year's MegaRun, which spun up 1.21 petaflops for an 18-hour workload.
While Gojira wouldn't quite make it into the Top 50 supers in the world, it only cost a little over US$5,500. ®