Quantifying NUMA and contention effects in multi-GPU systems

Kyle Spafford, Jeremy S. Meredith, Jeffrey S. Vetter

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

4 Scopus citations

Abstract

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.

Original languageEnglish
Title of host publicationProceedings of the 4th Workshop on General Purpose Processing on Graphics Processing Units, GPGPU-4
DOIs
StatePublished - 2011
Event4th Workshop on General Purpose Processing on Graphics Processing Units, GPGPU-4 - Newport Beach, CA, United States
Duration: Mar 5 2011Mar 5 2011

Publication series

NameACM International Conference Proceeding Series

Conference

Conference4th Workshop on General Purpose Processing on Graphics Processing Units, GPGPU-4
Country/TerritoryUnited States
CityNewport Beach, CA
Period03/5/1103/5/11

Keywords

  • GPGPU
  • benchmarking
  • graphics processors
  • performance

Fingerprint

Dive into the research topics of 'Quantifying NUMA and contention effects in multi-GPU systems'. Together they form a unique fingerprint.

Cite this