What I wonder is: do the sparse matrix routines of CHOMOD and UMFPACK take advantage of MKL
behind the scenes? Enabling or disabling MKL (using MKL) made absolutely no difference when calling cholesky and lu.
1 Like
Correction: there is a measurable difference in runtimes after all. But a minor one (~20%).
Measured on Surface Pro 7, running Windows 10.