Intel® Distribution of OpenVINO™ Toolkit
Community support and discussions about the Intel® Distribution of OpenVINO™ toolkit, OpenCV, and all things computer vision-related on Intel® platforms.
5959 Discussions

I can not confirm scaling in Pardiso with OpenMP



I am working on using Pardiso with OpenMP. But, I could not obtain speed-up.

I use Parallel_studio_xe_2017 update 8, for linux.


Based on the sample code, pardiso_sym_c.c, I developed the FEM code for 1D rod extension. The number of equations is 100,000,000 and the number of nonzero elements is 199,999,999 (0.000002%). And the matrix is symmetric.


I used the makefile provided by MKL example for compiling and running my code.


The summary of the results is as follows.

OMP_NUM_THREADS=1 :: (reordering phase, factorization phase, solution phase, total) = (147 s, 27 s, 11 s, 185 s)

OMP_NUM_THREADS=2 :: (reordering phase, factorization phase, solution phase, total) = (148 s, 36 s, 47 s, 231 s)

OMP_NUM_THREADS=4 :: (reordering phase, factorization phase, solution phase, total) = (148 s, 41 s, 53 s, 242 s)

OMP_NUM_THREADS=8 :: (reordering phase, factorization phase, solution phase, total) = (152 s, 33 s, 53 s, 238 s)

As you can see, I could not obtain speed-up.


Though I have searched for references, I haven't yet found good information.

Could you give me any advice?


I attached my code, the detailed results, and the speculation of my workstation. 






0 Kudos
0 Replies