Advanced

Accelerating crystal plasticity simulations using GPU multiprocessors

Mellbin, Ylva LU ; Hallberg, Håkan LU and Ristinmaa, Matti LU (2014) In International Journal for Numerical Methods in Engineering 100(2). p.111-135
Abstract
Crystal plasticity models are often used to model the deformation behavior of polycrystalline materials. One major drawback with such models is that they are computationally very demanding. Adopting the common Taylor assumption requires calculation of the response of several hundreds of individual grains to obtain the stress in a single integration point in the overlying FEM structure. However, a large part of the operations can be executed in parallel to reduce the computation time. One emerging technology for running massively parallel computations without having to rely on the availability of large computer clusters is to port the parallel parts of the calculations to a graphical processing unit (GPU). GPUs are designed to handle vast... (More)
Crystal plasticity models are often used to model the deformation behavior of polycrystalline materials. One major drawback with such models is that they are computationally very demanding. Adopting the common Taylor assumption requires calculation of the response of several hundreds of individual grains to obtain the stress in a single integration point in the overlying FEM structure. However, a large part of the operations can be executed in parallel to reduce the computation time. One emerging technology for running massively parallel computations without having to rely on the availability of large computer clusters is to port the parallel parts of the calculations to a graphical processing unit (GPU). GPUs are designed to handle vast numbers of floating point operations in parallel. In the present work, different strategies for the numerical implementation of crystal plasticity are investigated as well as a number of approaches to parallelization of the program execution. It is identified that a major concern is the limited amount of memory available on the GPU. However, significant reductions in computational time – up to 100 times speedup – are achieved in the present study, and possible also on a standard desktop computer equipped with a GPU. (Less)
Please use this url to cite or link to this publication:
author
organization
publishing date
type
Contribution to journal
publication status
published
subject
keywords
Crystal plasticity, Graphics processing unit, CUDA, GPGPU, Parallelization
in
International Journal for Numerical Methods in Engineering
volume
100
issue
2
pages
111 - 135
publisher
John Wiley & Sons
external identifiers
  • wos:000342668900002
  • scopus:84927633538
ISSN
1097-0207
DOI
10.1002/nme.4724
language
English
LU publication?
yes
id
aa1820fa-038c-4631-8e22-b2c756c9876b (old id 4645904)
date added to LUP
2014-09-15 10:16:38
date last changed
2017-09-17 03:23:31
@article{aa1820fa-038c-4631-8e22-b2c756c9876b,
  abstract     = {Crystal plasticity models are often used to model the deformation behavior of polycrystalline materials. One major drawback with such models is that they are computationally very demanding. Adopting the common Taylor assumption requires calculation of the response of several hundreds of individual grains to obtain the stress in a single integration point in the overlying FEM structure. However, a large part of the operations can be executed in parallel to reduce the computation time. One emerging technology for running massively parallel computations without having to rely on the availability of large computer clusters is to port the parallel parts of the calculations to a graphical processing unit (GPU). GPUs are designed to handle vast numbers of floating point operations in parallel. In the present work, different strategies for the numerical implementation of crystal plasticity are investigated as well as a number of approaches to parallelization of the program execution. It is identified that a major concern is the limited amount of memory available on the GPU. However, significant reductions in computational time – up to 100 times speedup – are achieved in the present study, and possible also on a standard desktop computer equipped with a GPU.},
  author       = {Mellbin, Ylva and Hallberg, Håkan and Ristinmaa, Matti},
  issn         = {1097-0207},
  keyword      = {Crystal plasticity,Graphics processing unit,CUDA,GPGPU,Parallelization},
  language     = {eng},
  number       = {2},
  pages        = {111--135},
  publisher    = {John Wiley & Sons},
  series       = {International Journal for Numerical Methods in Engineering},
  title        = {Accelerating crystal plasticity simulations using GPU multiprocessors},
  url          = {http://dx.doi.org/10.1002/nme.4724},
  volume       = {100},
  year         = {2014},
}