Quantifying NUMA and contention effects in multi-GPU systems
Oak Ridge National Laboratory, Oak Ridge, TN
Proceedings of the Fourth Workshop on General Purpose Processing on Graphics Processing Units, GPGPU-4, 2011
@inproceedings{spafford2011quantifying,
title={Quantifying NUMA and contention effects in multi-GPU systems},
author={Spafford, K. and Meredith, J.S. and Vetter, J.S.},
booktitle={Proceedings of the Fourth Workshop on General Purpose Processing on Graphics Processing Units},
pages={11},
year={2011},
organization={ACM}
}
As system architects strive for increased density and power efficiency, the traditional compute node is being augmented with an increasing number of graphics processing units (GPUs). The integration of multiple GPUs per node introduces complex performance phenomena including non-uniform memory access (NUMA) and contention for shared system resources. Utilizing the Keeneland system, this paper quantifies these effects and presents some guidance on programming strategies to maximize performance in multi-GPU environments.
September 23, 2011 by hgpu
