Intel® oneAPI Math Kernel Library
Ask questions and share information with other developers who use Intel® Math Kernel Library.

PARDISO

Pervane_Pishi
Beginner
863 Views
Hi dear friends,
I have a problem in PARDISO solver in MKL. I want exchange my solver with PARDISO to make the code faster, but sadly the factorization takes a long time. the old multigrid solver is not parallel, so i guess that PARDISO should be faster,but it is not. Do youhave any idea why it is not fast?
I would really appreciate your help.
Regards.
0 Kudos
4 Replies
Alexander_K_Intel2
863 Views

Hi,

We would like to help you but we need to have some details about size and type of your matrix, linkline, and processor and so on. Or could you provide testcase with PARDISO to reproduce low performance of this solver on our side to investigate problem deeper?

With best regards,

Alexander Kalinkin


0 Kudos
Gennady_F_Intel
Moderator
863 Views
also it would be helpful to have the same matrixes as you used while you encountered  this problem. 
Which mkl version you use? and 
What processor you're using while running that code?
regards
0 Kudos
sh_t_
Beginner
863 Views
Hi dear friend, I am using a sparse matrix with size of N=898560 and number of non zeros=3546432. i use CSR format for matrix storage. I want to use iterative format of PARDISO, but sadly i think i have implemented it wrongly. I want to know for using iterative case of pardiso do u have any example that which phases of solution should i use? and iterative with preconditioning, which values iparm and dparm should have. I am so confused sadly. I am not advance user of pardiso so i have no idea which values should i put for iparm. I do not want direct solver i want iterative with conjugate gradient preconditioner. BR,
0 Kudos
sh_t_
Beginner
863 Views
Hi again, I use MKL 10.2 version with 4 core.
0 Kudos
Reply