I have made inquiries before about the fact that Pardiso, though a fantastic performance solver, does not make performance use of multi-core architectures. This is verifiable by also looking at the Pardiso manual from the University in Basel Switzerland. Could someone please tell me if we get the new 4.0 version that O. Schenk and his team developed, or is Intel doing their own updates with an older version obtained from Basel ? I actually dont care how it gets done, but I need to know the answer to this question because my behind is on the line at my company because I promised them large performance improvements with investing in multi-core processors.
John, please take into acount that 4.0 of Olaf's PARDISO is not backward comatible with version 3.3 ( and with MKL's version ). into the Release Notes stated the comparision result of PARDISO v4.0 vs MKL PARDISO 10.2 for symmetric indefinite matrices. I've checked it and found that for some cases v4.0 faster then MKL PARDISO but produces the different results! I am sure in mkl's version but for version 4.0 need to do more checks. petr