Parallel integration of planetary systems on GPUs

Jianwei Gao, Eric B. Ford, Jörg Peters

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Direct numerical simulations of large numbers of planetary systems are key to the study extrasolar planetary systems. We investigate the potential of Graphics Processing Units (GPUs) to dramatically accelerate such calculations and enable new types of analyses. Here we report on a CUDA implementation of two different integrators, Verlet and Bulirsch-Stoer; that is, we are contrasting the relative efficiency of a low-memory-footprint, small-time-steps integrator with a more complex but highly stable integrator for best use of GPU resources. We verify that GPUs can serve as highly efficient dedicated parallel computing co-processors of the CPU host: the wall-clock time of integrating a large number of systems is reduced by two orders of magnitude compared to a comparable implementation we used so far on a modern CPU. We verify further that it is not only possible to port a complex integration algorithm like Bulirsch-Stoer to the restrictive GPU environment but that it can outperform the simpler Verlet integrator.

Original languageEnglish (US)
Title of host publicationProceedings of the 46th Annual Southeast Regional Conference on XX, ACM-SE 46
Pages7-12
Number of pages6
DOIs
StatePublished - Dec 1 2008
Event46th Annual Southeast Regional Conference on XX, ACM-SE 46 - Auburn, AL, United States
Duration: Mar 28 2009Mar 29 2009

Other

Other46th Annual Southeast Regional Conference on XX, ACM-SE 46
CountryUnited States
CityAuburn, AL
Period3/28/093/29/09

Fingerprint

Program processors
Direct numerical simulation
Parallel processing systems
Clocks
Data storage equipment
Graphics processing unit
Coprocessor

All Science Journal Classification (ASJC) codes

  • Computational Theory and Mathematics
  • Computer Science Applications
  • Hardware and Architecture
  • Software

Cite this

Gao, J., Ford, E. B., & Peters, J. (2008). Parallel integration of planetary systems on GPUs. In Proceedings of the 46th Annual Southeast Regional Conference on XX, ACM-SE 46 (pp. 7-12). [1593123] https://doi.org/10.1145/1593105.1593123
Gao, Jianwei ; Ford, Eric B. ; Peters, Jörg. / Parallel integration of planetary systems on GPUs. Proceedings of the 46th Annual Southeast Regional Conference on XX, ACM-SE 46. 2008. pp. 7-12
@inproceedings{687d38e5931148a8a07b584540bf7f07,
title = "Parallel integration of planetary systems on GPUs",
abstract = "Direct numerical simulations of large numbers of planetary systems are key to the study extrasolar planetary systems. We investigate the potential of Graphics Processing Units (GPUs) to dramatically accelerate such calculations and enable new types of analyses. Here we report on a CUDA implementation of two different integrators, Verlet and Bulirsch-Stoer; that is, we are contrasting the relative efficiency of a low-memory-footprint, small-time-steps integrator with a more complex but highly stable integrator for best use of GPU resources. We verify that GPUs can serve as highly efficient dedicated parallel computing co-processors of the CPU host: the wall-clock time of integrating a large number of systems is reduced by two orders of magnitude compared to a comparable implementation we used so far on a modern CPU. We verify further that it is not only possible to port a complex integration algorithm like Bulirsch-Stoer to the restrictive GPU environment but that it can outperform the simpler Verlet integrator.",
author = "Jianwei Gao and Ford, {Eric B.} and J{\"o}rg Peters",
year = "2008",
month = "12",
day = "1",
doi = "10.1145/1593105.1593123",
language = "English (US)",
isbn = "9781605581057",
pages = "7--12",
booktitle = "Proceedings of the 46th Annual Southeast Regional Conference on XX, ACM-SE 46",

}

Gao, J, Ford, EB & Peters, J 2008, Parallel integration of planetary systems on GPUs. in Proceedings of the 46th Annual Southeast Regional Conference on XX, ACM-SE 46., 1593123, pp. 7-12, 46th Annual Southeast Regional Conference on XX, ACM-SE 46, Auburn, AL, United States, 3/28/09. https://doi.org/10.1145/1593105.1593123

Parallel integration of planetary systems on GPUs. / Gao, Jianwei; Ford, Eric B.; Peters, Jörg.

Proceedings of the 46th Annual Southeast Regional Conference on XX, ACM-SE 46. 2008. p. 7-12 1593123.

Research output: Chapter in Book/Report/Conference proceedingConference contribution

TY - GEN

T1 - Parallel integration of planetary systems on GPUs

AU - Gao, Jianwei

AU - Ford, Eric B.

AU - Peters, Jörg

PY - 2008/12/1

Y1 - 2008/12/1

N2 - Direct numerical simulations of large numbers of planetary systems are key to the study extrasolar planetary systems. We investigate the potential of Graphics Processing Units (GPUs) to dramatically accelerate such calculations and enable new types of analyses. Here we report on a CUDA implementation of two different integrators, Verlet and Bulirsch-Stoer; that is, we are contrasting the relative efficiency of a low-memory-footprint, small-time-steps integrator with a more complex but highly stable integrator for best use of GPU resources. We verify that GPUs can serve as highly efficient dedicated parallel computing co-processors of the CPU host: the wall-clock time of integrating a large number of systems is reduced by two orders of magnitude compared to a comparable implementation we used so far on a modern CPU. We verify further that it is not only possible to port a complex integration algorithm like Bulirsch-Stoer to the restrictive GPU environment but that it can outperform the simpler Verlet integrator.

AB - Direct numerical simulations of large numbers of planetary systems are key to the study extrasolar planetary systems. We investigate the potential of Graphics Processing Units (GPUs) to dramatically accelerate such calculations and enable new types of analyses. Here we report on a CUDA implementation of two different integrators, Verlet and Bulirsch-Stoer; that is, we are contrasting the relative efficiency of a low-memory-footprint, small-time-steps integrator with a more complex but highly stable integrator for best use of GPU resources. We verify that GPUs can serve as highly efficient dedicated parallel computing co-processors of the CPU host: the wall-clock time of integrating a large number of systems is reduced by two orders of magnitude compared to a comparable implementation we used so far on a modern CPU. We verify further that it is not only possible to port a complex integration algorithm like Bulirsch-Stoer to the restrictive GPU environment but that it can outperform the simpler Verlet integrator.

UR - http://www.scopus.com/inward/record.url?scp=70449842981&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=70449842981&partnerID=8YFLogxK

U2 - 10.1145/1593105.1593123

DO - 10.1145/1593105.1593123

M3 - Conference contribution

AN - SCOPUS:70449842981

SN - 9781605581057

SP - 7

EP - 12

BT - Proceedings of the 46th Annual Southeast Regional Conference on XX, ACM-SE 46

ER -

Gao J, Ford EB, Peters J. Parallel integration of planetary systems on GPUs. In Proceedings of the 46th Annual Southeast Regional Conference on XX, ACM-SE 46. 2008. p. 7-12. 1593123 https://doi.org/10.1145/1593105.1593123