Nivolet Breaks Computing Records at Pratt
The research group of Guglielmo Scovazzi, associate professor of civil and environmental engineering, has just completed the deployment of the fastest computer cluster at Pratt.
Supported by a Defense University Research Instrumentation Program (DURIP) grant from the Army Research Office (ARO) and the Pratt School of Engineering, the new $300,000 setup dubbed Nivolet will help Scovazzi drastically cut down computing times for his research.
“Nivolet is the name of a mountain pass near Mount Grand Paradiso, in my homeland, Piemonte, Italy. It literally means ‘among the clouds,’” explains Scovazzi.
“What is impressive about this new cluster is that it is more powerful that the NASA Ames supercomputer I was using towards the end of my doctorate work in 2002. High performance computing has accomplished a massive jump in speed in a relatively short amount of time, while drastically reducing energy consumption and the trace on the environment.”
Scovazzi’s research focuses on finite element and advanced numerical methods for computational fluid and solid mechanics. In other words, he seeks to improve the accuracy of simulations of complex physical problems, such as subsurface flows in porous media, interactions between turbulent flows and structures in naval and aerospace applications, and geomechanics instabilities of the earth’s crust. These types of problems require splitting the overall computational effort among many powerful computer cores that must work in unison, in a synchronized divide-and-conquer approach.
Nivolet is a cluster of 32 nodes, each of which is equipped with 24 Intel Xeon cores, and connected through the industry-leading Mellanox SwitchX-2 Infiniband network to take full advantage of parallel computing for partial differential equation problems. The cluster also includes 13 Nvidia Tesla GPUs to further boost computing capabilities in dense linear algebra calculations.
“My research group is very excited about Nivolet because it will boost our research and its impact on complex computational mechanics problems,” says Scovazzi. “We’re grateful for the support from the ARO Applied Mathematics Program managed by Dr. Joe Myers. Also, I would particularly like to thank Jim Daigle, Jeremy Thornhill and Ron Stubbs of Pratt OIT who helped getting the cluster up and running in record time. We feel blessed in my team to have their help.”