In-Core LU-Decomposition of Symmetrical Dense MoM Matrix in WIPL-D Multi-GPU Solver

摘要

Acceleration of in-core LU decomposition of symmetrical dense MoM matrices by using multiple GPUs in parallel is presented in this paper. Memory limitations of GPUs are overcome by using block LU decomposition, where the entire system matrix is stored in CPU RAM, while only processed blocks are stored in GPU VRAM. The presented algorithm for LU decomposition enables highly efficient utilization of an arbitrary number of GPUs. Comparison of performance of up to 4 GTX 680 GPUs and up to 4 GTX 1080 Ti GPUs is shown. Presented results show that a symmetrical dense MoM matrix with 100 000 complex unknowns in single precision can be LU decomposed in about 3.5 minutes, on a personal computer equipped with 4 GTX 1080 Ti GPUs.