I’m trying to use Cholesky to solver symmetric sparse matrix. I have tested my matrix on both cusolverSpDcsrlsvchol and the low level Cholesky using codes in samples. However, both of them use much more time to solve the matrix than MKL PARDISO library on 8 CPU cores. The reordering and factorization methods are the same.
My question is:
Is there a way or some settings I can take to further accelerate the Cholesky on GPU? Now, my settings follow the sample code.
The GPU I use is Tesla V100 and CUDA is 10.1.