You need to call the Multiprocessing.pool inside of the function and have it map directly to the numpy matrix inversion.

But, if the task is not big enough I cant guarantee that it will speed it up. The independent cost of setting up the process might outdo the singular execution of each.

"Embarassingly Parrellel" is a term used for GPU multiprocessing. Currently python doesnt support native gpu acceleration.