opencl (if you've not heard of it, it's a way to program graphics cards - very like cuda, but cross-platform). we had some numerical code in matlab that was ta

opencl (if you've not heard of it, it's a way to program graphics cards - very like cuda, but cross-platform). we had some numerical code in matlab that was taking too long to run. i managed to get a speed-up of 80x by moving it to a gpu - and the gpu in question is not even the latest generation, but something we picked up on newegg for less than $200. and i suspect that's just for starters - i haven't even profiled it yet to see how to make it faster. using opencl isn't that much harder than c. obviously you need to understand a little about how gpus work, and there are all the same issues about allocating memory, segmentation faults, etc. but the compiler and runtime are quite solid, and the docs more than adequate. now that i am past the worst of the learning curve i'm looking around for more slow code, because it's an easy way to get more speed. also, django, but others have mentioned that (it's such a breath of fresh air if you're used to using java...). next year, i hope to play with llvm some (although it will have to be on my own time - can't see how to do that at work...)