Abstract
To farther accelerate the three-dimensional two-scale stress analyses of composite material with iterative conjugate gradient method and voxel finite element method, we propose a code implementation for GPU. In the framework, several specialized algorithms can be employed; i.e., element-by-element technique utilizing explicit element-to-nodes connectivity, FFT pre-conditioner accelerating the convergence of the iteration, and so on. These algorithms are implemented for GPU computations by CUDA to gain more speed compared to PC computations. Numerical experiments have exhibited that the GPU computations are several times faster than the PC computations.
