For many computationally intensive tasks, exacting precision is not necessary for every step of the entire task to obtain a suitably precise result. The alternative is mixed-precision computing: using high precision where it matters, using low precision where it doesn’t as much. Now, a research team led by the King Abdullah University of Science and Technology (KAUST) has applied mixed-precision computing to drastically reduce the computational load of large geospatial dataset modeling.
“For decades, modeling of environmental data relied on double-precision arithmetic to predict missing data,” explained Sameh Abdulah, a research scientist at KAUST and first author of the paper. “Today, there is high-performance computing hardware that can run single- and half-precision arithmetic with a speedup of 16 and 32 times compared with double-precision arithmetic. To take advantage of this, we propose a three-precision framework that can exploit the acceleration of lower precision while maintaining accuracy by using double-precision arithmetic for vital information.”
The framework operates using a runtime called PaRSEC (Parallel Runtime Scheduling and Execution Controller) developed by a team at the University of Tennessee Knoxville (whose researchers also contributed to this work). PaRSEC, “a generic framework for architecture aware scheduling and management of micro-tasks on distributed many-core heterogeneous architectures,” allowed the researchers to orchestrate the mixed-precision approach across the hardware (which KAUST only specified as a “implemented on a high-performance computing system based on highly parallelized … GPUs”).
With PaRSEC in-hand, the researchers applied the framework to geostatistical modeling of a massive dataset, using statistical relationships among the data to select certain weakly correlated data for lower (single- or half-) precision, reserving double-precision analysis for more strongly correlated data.
“The main goal of this project is to leverage the recent parallel linear algebra algorithms developed by KAUST’s Extreme Computing Research Center to scale up geospatial statistics applications on leading-edge parallel architectures,” Abdulah said. “We have shown that we can achieve significant speedup compared to full double-precision arithmetic modeling while preserving the parameter estimations and prediction accuracy to meet the application requirements. … Next, we intend to integrate approximations with mixed precision to further reduce memory footprint and shorten calculation time.”
About the research
The research discussed in this article was published as “Accelerating Geostatistical Modeling and Prediction With Mixed-Precision Computations: A High-Productivity Approach with PaRSEC” in the May 2021 issue of IEEE Transactions on Parallel and Distributed Systems. It was written by Sameh Abdulah, Qinglei Cao, Yu Pei, George Bosilca, Jack Dongarra, Marc M. Genton, David Keyes, Hatem Ltaief and Ying Sun. To read it, click here.