Three orders of magnitude improved efficiency with high-performance spectral crystal plasticity on GPU platforms

SUMMARY We study efficient numerical implementations of crystal plasticity in the spectral representation, with emphasis on high-performance computational aspects of the simulation. For illustrative purposes, we apply this approach to a Taylor homogenization model of fcc poly-crystalline materials and show that the spectral representation of crystal plasticity is ideal for parallel implementations aimed at next-generation large-scale microstructure-sensitive simulations of material deformation. We find that multi-thread parallelizations of the algorithm provide two orders of magnitude acceleration of the calculation, whereas graphics processing unit-based computing solutions allow for three orders of magnitude speedup factors over the conventional model. Copyright © 2014 John Wiley & Sons, Ltd.