A new mathematical solver enables the graphics cards found in gaming computers to solve computationally intensive mathematical problems.
Researchers at the King Abdullah University of Science and Technology's (KAUST) Extreme Computing Research Center in Saudi Arabia modified a graphics-processing unit (GPU) to include a more efficient solver. Acceleration of the solver could considerably reduce the execution time and energy required to solve the problem.
GPUs are more energy efficient compared with standard high-performance processors because they eliminate much of the hardware used by standard processors to executive general-purpose code. However, GPUs' supporting software is immature, so researchers aimed to make a more efficient system that maximizes the trade-off between the number of processors and the memory available to temporarily store the data.
A solver scheme was designed to operate directly on data without making an extra copy, meaning a system twice as large can be stored in the same amount of memory. The KAUST researchers also redesigned the way in which simultaneous equations are performed, implementing a triangular matrix-matrix multiplication method that achieves up to an eightfold acceleration compared with the speed of existing systems.
The advanced solver will be integrated into the next software library for NVIDIA GPUs.
From KAUST Discovery
View Full Article
Abstracts Copyright © 2017 Information Inc., Bethesda, Maryland, USA