Intel® Math Kernel Library

Difference of calculcation result between i7 and xeon cpu


I make the numerical calculation program and run two computers. (One is used i7 cpu, others Xeon cpu.)

The numerical result is slightly different. (input data is same).

My program is very sensitive to numerical difference, so i confused which one is correct.

Why the differece is occured?


the program is complied same computer under environment

Complier : MS VS2010 C++ Complier

MKL library Version :

Linked library : mkl_core.lib, mkl_intel_lp64.lib, mkl_intel_thread.lib, libiomp5md.lib


MKL Pardiso (version 11.2.3): wrong output of phase 331 with multiple rhs and Schur complement enabled


I recently started using MKL_PARDISO. I noticed that phase 331 gives the wrong result if you want to solve for multiple right hand sides using the Schur complement feature.

Attached a code to reproduce the problem. I just copied the example you provide with mkl distribution for the Schur complement and added multiple rhs.

I'm using composer_xe_2015.3.187, with MKL 11.2.3


Memory allocation issues with `cluster_sparse_solver` in MKL 11.3 with distributed matrix input

Calling `cluster_sparse_sovler` with `-DMKL_ILP64`, where the input matrix is distributed with `iparm[39]=2`, produces the following:

*** Error in PARDISO  (     insufficient_memory) error_num= 1                                                                           
*** Error in PARDISO memory allocation: MATCHING_REORDERING_DATA, allocation of 1 bytes failed                                          
total memory wanted here: 75743 kbyte

 There is no effect from the number of nodes used for the MPI job.

I need help. Wrong results when i use dgels, ubuntu 14.04+mkl 11.0


I want to use the MKL library for solving a system of overdetermined linear equations.For this I use the dgels LAPACK function, which is provided by MKL library. In my particular problem (matrix of 4800 rows and 81 columns) the results are incorrect: Both, the QR factorization and the solution are wrong. But seems to work good for toy matrices (6x4 dgels matrix example in mkl documentation).

MKL gesv memory usage


Testing LAPACK_zgesv, 

I saw that it allocates a large amount of memory, in particular this can crash my rogram when for example a matrix 5000x5000 is passed.
Is there any alternative solution or option in order to limitate this overuse of memory?

I'm using MKL 11.3.01 in C# if it can be helpfull.




Intel® Math Kernel Library abonnieren