Publication detail

Increasing Double Precision Throughput on NVIDIA Maxwell GPUs

POLOK, L. SMRŽ, P.

Original Title

Increasing Double Precision Throughput on NVIDIA Maxwell GPUs

Type

conference paper

Language

English

Original 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.

Keywords

double precision calculation, multiple precision arithmetics, GPGPU

Authors

POLOK, L.; SMRŽ, P.

Released

3. 4. 2016

Publisher

Association for Computing Machinery

Location

Pasadena / Los Angeles

ISBN

978-1-5108-2318-1

Book

Proceedings of the 24th High Performance Computing Symposium

Pages from

146

Pages to

153

Pages count

8

URL

BibTex

@inproceedings{BUT130935,
  author="Lukáš {Polok} and Pavel {Smrž}",
  title="Increasing Double Precision Throughput on NVIDIA Maxwell GPUs",
  booktitle="Proceedings of the 24th High Performance Computing Symposium",
  year="2016",
  pages="146--153",
  publisher="Association for Computing Machinery",
  address="Pasadena / Los Angeles",
  isbn="978-1-5108-2318-1",
  url="http://dl.acm.org/citation.cfm?id=2972989"
}