Since multicore processors were first developed and marketed, software developers have been working to take advantage of their parallel computing capabilities. Even if a program is not written to utilize multiple cores, the user still sees a benefit as the remaining cores are free to run other programs. Now researchers at the Stanford School of Engineering have successfully run a program across one million computing cores to model computational fluid dynamics.
The supercomputer named Sequoia is located at Lawrence Livermore National Laboratories and has 1,572,864 computing cores with 1.6 petabytes of memory. Using every core is exceedingly difficult though because just as the computation is scaled up to the multiple cores, so are any errors in the computation. Even something small that would never crash a simpler computer can cause a failure, but after weeks of working out the bugs, the researchers were able to watch the performance scale past the million-core threshold.
The model the researchers were running was a noise simulation from a jet engine, which is important for finding ways to keep the noise down. Thanks to the million cores, the simulation was completed dramatically quicker than with fewer processors.
Source: Stanford School of Engineering