[Ipopt] Multiple Cores

[Tony Kelman]

Mosy,

The computational bottleneck in Ipopt is typically the linear system solution. There are a variety of sparse linear solvers you can use with Ipopt, several of which (PARDISO, WSMP, and HSL MA86 and MA97) are capable of parallel solution using multithreading. You can also get some parallelism from an optimized BLAS implementation like Intel MKL or OpenBLAS, but in my experience unless your problem is very dense you are better off going for parallelism in the sparse solver than in BLAS.

The 4 mentioned parallel linear solvers are available under different licenses. You’ll have to look into obtaining one or more of them, and compile Ipopt with the appropriate options (like --with-pardiso=”...” or --with-wsmp=”...”, see http://www.coin-or.org/Ipopt/documentation/node19.html#ExpertInstall).

-Tony
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/ipopt/attachments/20130510/495d603a/attachment.html>