I'm using Intel MKL 10.3 update 1.
When I cmpare time to solve Ax=b with Pardiso, I get slower time with multiple right hand side than in single rhs.
I made many tests with different matrix size and number of non zero values and I get always same behavior.
Matrix is symmetrical
Matrix and vector B are fill with complex number
Matrix size : 10000x10000
Number of non zero value per row : 3
All non zero values are on the diagonal and near the diagonal
Vector B is fill with 1+j0 at each index. (to simplify the test)
Reordoring and Factorisation are runned only 1 time.
Time to solve 10000 times the system with 1 rhs : 4.485 sec.
Time to solve 10000/8 times the system with 8 rhs: 15.89 sec.
Time to solve 10000/32 times the system with 32 rhs: 7.672 sec.
Time to solve 10000/128 times the system with 128 rhs: 6.89 sec.
Is it normal that the faster result is always wit 1rhs no matter the matrix size, the number of non zero and values in B vector ?
I'm on windows XP, visual studio 2005 and 2010
Intel Core 2 Quad CPU Q9550 @ 2.83GHz 3.25 GB RAM
