Opening Up the Accelerator Advantage

By Tiffany Trader

Researchers at Georgia Institute of Technology and University of Southern California will receive nearly $2 million in federal funding for the creation of tools that will help developers exploit hardware accelerators in a cost-effective and power-efficient manner. The purpose of this three-year NSF grant is to bring formerly niche supercomputing capabilities into the hands of a more general audience to help them achieve high-performance for applications that were previously deemed hard to optimize. The project will involve the use of tablets, smart phones and other Internet-era devices, according to David Bader, the lead principal investigator.

“We want to take science that used to be only available to elite scientists and bring that to everyone around the planet,” said Bader, a professor in Georgia Tech’s School of Computational Science and Engineering and executive director for High Performance Computing. “We are bringing supercomputing to the masses.”

An article at Georgia Tech details the project’s two main focal points: XScala and XBazaar. XScala is a software framework for developing efficient accelerator kernels. The framework includes a number of design time and run-time performance optimization tools designed to handle data-intensive kernels, those bound by data movement. Examples include large dictionary string matching, dynamic programming, graph theory, and sparse matrix computations — commonly associated with biology, network security, and the social sciences workloads.

Researchers will work on different types of optimizations over the three-year period. Power efficiency will receive a lot of attention, as will security and social network analysis. The researchers expect that algorithms developed from these areas can then be applied to the graph analytics domain.

The second focus involves a public software repository and forum, called XBazaar. This is similar to an app store, like on the iPhone. “XBazaar will serve as a one-stop shop for high-performance algorithms and software for multi-core and many-core processors,” according to the announcement.

It’s a place where developers can go to share best practices for using accelerators. It’s open source so anyone can use it to develop their own high-performance applications, even commercial applications. The idea is to create a tool that can benefit academic researchers and industry.

Besides Bader, the team includes co-PIs: Viktor Prasanna, a professor of electrical engineering and professor of computer science at the University of Southern California; Rich Vuduc, an associate professor in the School of Computational Science and Engineering at Georgia Tech; and Jason Riedy, a research scientist.

David A. Bader
David A. Bader
Distinguished Professor and Director of the Institute for Data Science

David A. Bader is a Distinguished Professor in the Department of Computer Science at New Jersey Institute of Technology.