Previously, I posted about the enormous performance gain when using NumPy for repeated calculations – especially when the calculations are repeated within a for loop.
There is a great blog here that profiles the benefits of native Python vs. Numpy vs. Cython. Really good read.
And similarly, an experiment here that compares Python, NumPy, Fortran and Java. Not surprisingly NumPy far outstrips Java and Python, but surprisingly (to me at least), it is on par with Fortran and MAtlab for some of their experiments. Particularly interesting is Problem 3, which uses an iterative algorithm to solve a Laplace equation. This could be really useful reference for the WBGT relaxation algorithm.