Intel® oneAPI Math Kernel Library
Ask questions and share information with other developers who use Intel® Math Kernel Library.
6956 Discussions

Stack overflow in Pardiso solver Error

Hoang__Dat
Beginner
862 Views

Dear all,

I would like to ask for your help to solve this error when I used Pardiso for my Finite element analysis.

I am using Intel® Parallel Studio XE 2017 update 6 for Windows with Intel Fortran compiler integrated in Visual Studio 2015. My laptop has 8 Gb RAM, 4 processors 2.0 GHz. 

I do not know why when I solved the equation Ax=b where A size [90,000 x 90,000] with the setting of Pardiso solver as the attached picture 3, there was an error relating to stack overflow as the attached pictures 1&2. When running only 45% of RAM memory was used.

Please help me find the reason and the way to fix this problem. 

Thank you very much.

 

0 Kudos
1 Solution
Gennady_F_Intel
Moderator
862 Views

Do you see the problem at the phase 11, 22 or 33?  

Would be much better if you will give us the reproducer and your matrix to check the problem on our side.  

Meantime you may try to take the latest MKL 2018 u3, or try to use OOC version ( iparm[59]=2) .

View solution in original post

0 Kudos
4 Replies
Gennady_F_Intel
Moderator
863 Views

Do you see the problem at the phase 11, 22 or 33?  

Would be much better if you will give us the reproducer and your matrix to check the problem on our side.  

Meantime you may try to take the latest MKL 2018 u3, or try to use OOC version ( iparm[59]=2) .

0 Kudos
Hoang__Dat
Beginner
862 Views

Thank you very much.

I have solved my problem. My error is that I load too much dummy matrice which make the subroutine of boundary overload. I reduced the size of the dummy matrice.

By the way, I would like to ask about the capacity of Pardiso solver.

I want to solve the Ax=b where A size is [4.5 millions x 4.5 millions]. How large is the memory of RAM needed for Pardiso to solve that equation? My PC has 16 cores.

I look forward to receiving your help.

Yours sincerely,

0 Kudos
Gennady_F_Intel
Moderator
862 Views

Actually there are no reliable methods to predict the exact #nnz ( therefore consumed memory). We recommend to run the reordering phase ( 11)  and see the #of non zeroes elements in factors by printing iparm[17] parameters.

0 Kudos
Hoang__Dat
Beginner
862 Views

Dear Mr. Gennady,

Thank you very much. I have already understood this issue. 

0 Kudos
Reply