topic Matlab and MKL PARDISO comparison in Intel® oneAPI Math Kernel Library

Matlab and MKL PARDISO comparison

programmer85 — Fri, 05 Mar 2010 14:08:45 GMT

As long as I know Matlab take advantage of MKL.

[1] How can it happen that for the same unsymmetric sparse matrix (size of 20, 000 x 20, 000) that

In Matlab using A\\b and MKL PARDISO using steps ( analysis and symbolic factorization, numerical factorization forward and backward substitution including iterative refinement, termination to release all internal solver memory) I got about 300 times better performacne using Matlab than using PARDISO. ( I used the same CPU Intel i7 in tests).

???

[2] Are there any special options that could be applied to speed-up performance for PARDISO ?

Matlab and MKL PARDISO comparison

Gennady_F_Intel — Fri, 05 Mar 2010 15:37:58 GMT

adams,

yes, Matlab uses MKL inside. The performance problem that you bring to our attention is quite newfor us.How can we check it? Can you give us this unsymmetricmatrix for performing the result on our side?

What version of MKL you using? OS?

Was it in-core, hybrid or out-of-core modes?

Can you give the concrete performance numbers you got?

--Gennady

Matlab and MKL PARDISO comparison

programmer85 — Sat, 06 Mar 2010 09:58:47 GMT

send me a message with your mail and I will send you matrix with code and settings that I use

Matlab and MKL PARDISO comparison

Melvin_Robinson — Sat, 06 Mar 2010 21:29:28 GMT

Can you post the code and attach the file?

Matlab and MKL PARDISO comparison

Gennady_F_Intel — Sun, 07 Mar 2010 07:34:39 GMT

You can create the Private post and attach all data and source code which you want to share for us. The private post will guarantee that these data and source code will available only for you, me and my collegues from our ( MKL) team.

--Gennady