topic Re:How to get the information about improve speed ... in Intel® oneAPI Math Kernel Library

How to get the information about improve speed in cluster_sparse_solver distributed assembled matrix version

DongHee_B_ — Fri, 29 May 2020 08:00:15 GMT

Hi. I test to get solution about non-symmetric matrix Ax=B used cluster version direct sparse solver.

I used to MKL 2018 update 3 and the compiler is VS 2017.

I test to my application used cluster_sparse_solver both option centralized and distributed. (iparm[39])

And the solving time of distributed option solver is x3 ~ x5 slower than centralized option solver.

(The size of matrix about quarter million. Used same A and B, Ax = b result is same)

I tried the start and end domain number of distributed option solver is just divided number of process. (4 process, 0 ~ 30000, 30001~60000, ...)

Is there anything I did wrong? If not, what information should be used to divide the domain?

Application Note:

Gennady_F_Intel — Fri, 29 May 2020 10:49:56 GMT

Application Note: "Performance of the reordering step of the Parallel Direct Sparse Solver for Clusters Interface is slightly better for assembled format (CSR, iparm[39] = 0) than for distributed format (DCSR, iparm[39] > 0) for the same matrices, so if the matrix is assembled on one node do not distribute it before calling cluster_sparse_solver."

But 3...5x times is not what we expect to see. Did you check these numbers with the latest ( 2020) versions of intel mkl sparse solver for cluster?

Re:How to get the information about improve speed ...

Gennady_F_Intel — Fri, 26 Jun 2020 05:47:05 GMT

Have you had a chance to check these numbers with the latest ( 2020) versions of intel mkl sparse solver for cluster?