Publication Details

Increasing Double Precision Throughput on NVIDIA Maxwell GPUs

POLOK Lukáš and SMRŽ Pavel. Increasing Double Precision Throughput on NVIDIA Maxwell GPUs. In: Proceedings of the 24th High Performance Computing Symposium. Pasadena / Los Angeles: Association for Computing Machinery, 2016, pp. 146-153. ISBN 978-1-5108-2318-1. Available from: http://dl.acm.org/citation.cfm?id=2972989
Czech title
Zvyšování prostupnosti v dvojité přesnosti na grafických kartách Maxwell
Type
conference paper
Language
english
Authors
Polok Lukáš, Ing. (DCGM FIT BUT)
Smrž Pavel, doc. RNDr., Ph.D. (DCGM FIT BUT)
URL
Keywords

double precision calculation, multiple precision arithmetics, GPGPU

Abstract

This paper deals with the impact the architectural changes of modern GPUs have on their use in scientific computing. It particularly focuses on significant drops in the number of double precision functional units in NVIDIA Maxwell architecture. Proposed remedies of the potential negative impact on GPGPU applications that are based on multiple precision arithmetics are discussed. Two new algorithms for fast and precise multiplication and fused multiply add for double precision arithmetics emulation are also presented here. 

Using these methods, we were able to boost the double precision performance of NVIDIA GTX 980 Ti from 95 GFLOPS up to 286 GFLOPS. The proposed methods are applicable also to other GPUs.

Published
2016
Pages
146-153
Proceedings
Proceedings of the 24th High Performance Computing Symposium
Conference
24th High Performance Computing Symposium (HPC 2016), Pasadena, CA, US
ISBN
978-1-5108-2318-1
Publisher
Association for Computing Machinery
Place
Pasadena / Los Angeles, US
EID Scopus
2-s2.0-84977080503
BibTeX 
@INPROCEEDINGS{FITPUB11076,
   author = "Luk\'{a}\v{s} Polok and Pavel Smr\v{z}",
   title = "Increasing Double Precision Throughput on NVIDIA Maxwell GPUs",
   pages = "146--153",
   booktitle = "Proceedings of the 24th High Performance Computing Symposium",
   year = 2016,
   location = "Pasadena / Los Angeles, US",
   publisher = "Association for Computing Machinery",
   ISBN = "978-1-5108-2318-1",
   language = "english",
   url = "https://www.fit.vut.cz/research/publication/11076"
}
Back to top