George Biros, leader of the ICES Parallel Algorithms for Data Analysis and Simulation Group, has received a $750,000 grant from the National Science Foundation (NSF) to improve the energy efficiency of computing an important class of problems.
The three-year grant is part of the NSF’s Exploiting Parallelism and Scalability Program, which awarded approximately 20 awards this year to support “groundbreaking research leading to a new era of parallel computing.”
Co-investigators of the grant research are electrical and computer engineering professors Andreas Gerstlauer and Lizy K. John, as well as ICES researcher Robert van de Geijin.
The goal of the grant research is akin to increasing gas mileage in automobiles, according to the proposal. But instead of miles per gallon, Biros and others are aiming on increasing the computational performance per unit of power, or FLOPS per watt.
To do this, the team is surveying the energy efficiency of N-body and tree-based finite element method problems across current computer architectures. By applying this information, they plan to design new, efficiency-boosting algorithms and architectures and integrate them into application-specific integrated circuits for handling these specific problem types.
“The goal is to have energy efficient circuits,” said Biros. “Today’s machines are not as efficient because they sacrifice efficiency for generality.”
N-body and tree-based finite element methods are used to solve problems mathematical physics and statistical inference. Their application is relevant across computational science, says Biros, making improved computational efficiency an interest to scientists ranging from chemists to cosmologists.
In addition to conducting research, the scientists plan to use the grant as an educational opportunity. Three PhD students working on the research plan to integrate theoretical and practical research results into graduate and undergraduate courses they teach, as well as into the university’s Freshman Research Initiative.
“This will fill a crucial gap in cross-disciplinary education on system design, where there is an increasingly essential need to think across hardware/software and algorithm/architecture boundaries,” said the grant proposal.