Evaluating the Energy Efficiency of OpenCL-accelerated AutoDock Molecular Docking
Embedded Systems and Applications Group, Technische Universitat Darmstadt, Darmstadt, Germany
28th EUROMICRO International Conference on Parallel, Distributed and Network-Based Processing (PDP’20), 2020
@inproceedings{solis2020evaluating,
title={Evaluating the Energy Efficiency of OpenCL-accelerated AutoDock Molecular Docking},
author={Solis-Vasquez, Leonardo and Santos-Martins, Diogo and Koch, Andreas and Forli, Stefano},
booktitle={Submitted to the 28th Euromicro International Conference on Parallel, Distributed, and Network-based Processing (PDP). Submitted},
year={2020}
}
AUTODOCK is a molecular docking application that consists of a genetic algorithm coupled with the Solis-Wets localsearch method. Despite its wide usage, its power consumption on heterogeneous systems has not been evaluated extensively. In this work, we evaluate the energy efficiency of an OpenCL-accelerated version of AUTODOCK that, along with the traditional SolisWets method, newly incorporates the ADADELTA gradient-based local search. Executions on a Nvidia V100 GPU yielded energy efficiency improvements of up to 297x (Solis-Wets) and 137x (ADADELTA) with respect to the original AUTODOCK baseline.
March 1, 2020 by hgpu
