I use PARDISO to solve a symmetric sparse matrix with size n=400,000 and number of nonzero=1,100,000. When I set the number of thread to one, I can get the same answer every time. But when I set the number of threads higher than 1 and run it many times, the answers become different every time (and never same as the answer by one thread). Are these weird results related to the issues of condition number or something else?
By the way, I have tried both MKL 11.0.2 and 11.2.3.
Regarding PARDISO generates different results when different threads . in most of case, the behavior was expected as explained in the KB article: https://software.intel.com/en-us/articles/getting-reproducible-results-with-intel-mkl
And we also introduce the feature of Conditional Numerical Reproducibility (CNR) in https://software.intel.com/en-us/articles/introduction-to-the-conditional-numerical-reproducibility-....
Have you validated the result, are they acceptable? any way, in case of other problems, it would be great if you have test case for us to check in details.