[Coin-ipopt] Re: Vol 35, Issue 8 ... optimized blas

Fabian Wein fabian.wein at lse.e-technik.uni-erlangen.de
Wed Nov 14 04:19:41 EST 2007

Hi Olaf,

now, with the gcc 4.2 OpenMP version out. Shall one use the OpenMP
version of ACML
cuncurrently with the OpenMP version of PARDISO?

Thank you,

Fabian Wein

> I have just one comment on your email. It terms of performance it would be much
> better to use LAPACK from ACML/MKL or ATLAS instead of your own compiled Netlib
> distribution. The performance of all modern sparse direct solvers
> (MUMPS/PARDISO, not MA27) is typically a factor of 10 (!) times faster with
> optimized LAPACK libraries.

More information about the Coin-ipopt mailing list