July 27, 2012

Soft Serve Supercomputing

Robert Gelber

A team of researchers at the University of Edinburgh, Scotland has embarked on a palatable study. Using a supercomputer, scientists are simulating ice cream in hopes to improve the popular dessert’s texture and shelf life. NVIDIA reported on the project in a recent blog post.

One of the lead scientists noted that ice cream is a very complex substance. The multiple ingredients found in popular recipes react with each other in a variety of ways over time. To get a better understanding of how they interact, the group decided to perform a computer simulation.

The processing power required to study ice cream at the molecular level would take years for consumer-grade computers to accomplish. To get faster results, the team turned to the Edinburgh Parallel Computing Center (EPCC), where they researched the frozen treat on a 200K-core Cray supercomputer.

During the project, the team realized they were able to perform the same computations on a far smaller, GPU-accelerated cluster. The simulations were migrated from a 200-cabinet system to a 10-cabinet, GPU-accelerated Cray XK6. The smaller cluster had 936 Tesla GPUs, which helped the team complete their simulation two and a half times faster than with CPUs alone.

While ice cream is a tasty and unique substance to learn more about, the same simulations could apply to other soft materials. University of Edinburgh’s Alan Gray explained that paint, ketchup, yogurt, and hair products are just a few examples of items applicable to this research.