I used OpenCL several years ago, to integrate large systems of coupled partial differential equations on an AMD GPU (at the time rated as the world's fastest).

I used OpenCL several years ago, to integrate large systems of coupled partial differential equations on an AMD GPU (at the time rated as the world's fastest). Using the hardware efficiently required significant effort, and in the end, the result was no better than what I got from Intel's optimizing compiler and an overclocked Core i7 CPU, which had the additional (and for that application decisive) advantage of proper double precision support. Don't know how the latest hardware would stack up though.