Jacket is a nice addition to Matlab, which like any interpreted language is inherently slow. I've seen a 40x slow down for the same application versus a Fortran implementation.<p>In my view Matlab is great for testing your ideas, but for an industrial application you will translate your code to C/C++ or Fortran.<p>I'm curious to check how much speed you can achieve using Jacket versus a native C+CUDA code. I doubt Matlab+Jacket can run at the same speed as C+CUDA, but I will give it a try.