On increasing capabilities of GPU in-core solver applied to method of moments
Dusan P. Zoric, Dragan I. Olćan, Branko M. Kolundžija · 2012
We increase the computation and memory capabilities of GPU in-core solver by using block LU decomposition, RAM for storing the whole matrix and VRAM for storing currently used block(s) of data. Using up to 3 GPUs gives parallelization efficiency of up to 70%. The matrix of EM problems with 50 000 unknowns can be LU decomposed within 4 minutes on a PC with used GPUs.