• You need to use CUDA*[] functions explicitly in your code to be able to exploit your GPU. The built-in functions do not automagically detect if you want to use a GPU. – J. M. will be back soon Nov 8 '17 at 14:37
• Thanks. But as far as I can tell, I can use CUDAFunctionLoad[] to load a function I write in C, which can be anything, right? – A. Vieira Nov 8 '17 at 15:05
• That's what I thought. Thanks. I know what CUDA is and maybe it will be worthwhile some day, but for now Parallelize[] and ParallelizeDo[] plus a multi-core will do the job. – A. Vieira Nov 8 '17 at 16:50
• Usually CUDA helps when you have large matricies that need a lot of processing. Even better you have little matrices on the RAM and many lengthy calcuations possible on the GPU. The reason is the memory reading time. If you have to calculate e.g. a determinat CUDA would be rather slow. Usually re-thinking the code and partically Compile and parallelize it in Mathematica would give speed ups. With GPU would certainly need to add own functions to the once Mathematica provides. This also works but additional knowledge and effort is needed. – Eisbär Nov 10 '17 at 12:01