12982

On the Efficiency of CPU and Hybrid CPU-GPU Systems in Computational Biology Tasks

Nikolay Pydiura, Pavel Karpov, Yaroslav Blume
Institute of Food Biotechnology and Genomics, Natl. Academy of Sciences of Ukraine, Kyiv 04123, Ukraine
Computer Science and Applications, Volume 1, Number 1, pp. 48-59, 2014

@article{pydiura2014efficiency,

   title={On the Efficiency of CPU and Hybrid CPU-GPU Systems in Computational Biology Tasks},

   author={Pydiura, Nikolay and Karpov, Pavel and Blume, Yaroslav},

   journal={Computer Science},

   volume={1},

   number={1},

   pages={48–59},

   year={2014}

}

Download Download (PDF)   View View   Source Source   

906

views

The complexity and diversity of the computational biology tasks requires a deliberate approach to the computational resource management. We have analyzed the performance of the common CPU and hybrid CPU-GPU hardware configurations in molecular dynamics and homology modeling tasks. Our results show that on dual-processor nodes it is in overall more efficient to execute two jobs simultaneously on each CPU of the node rather than run these jobs consequently on both processors of the node. Analysis of the test data available in the literature and our own Gromacs molecular dynamics tests of hybrid acceleration with GPUs show at least 50% boost of the computing performance over CPU-only calculations. Thus, we suggest that any cluster node or workstation intended for scientific calculations should be preferably equipped with one or two graphics cards corresponding to the installed CPU. We suggest that consumer level graphics cards may be applicable in hybrid CPU-GPU systems for stochastic single-precision scientific calculations. NVidia GTX 780, GTX 780 Ti, and ATI R9 290, R9 290X graphics cards are the best choice for single-precision computing and GTX Titan is an unmatched choice for double-precision computing among consumer level graphics cards at present time.
No votes yet.
Please wait...

Recent source codes

* * *

* * *

HGPU group © 2010-2018 hgpu.org

All rights belong to the respective authors

Contact us: