Again i'll have to correct you on this - it is not native support. It is emulation. Emulation is almost always inefficient and buggy when trying to run hardware supported calculations on hardware which does not have those exact calculations supported. You would not get a very good result running CUDA on STREAM hardware (on the 4k series) because CUDA had things such as double floating point precision (i think that is correct, been a while since i looked into the fundamental differences), which stream and the 4k series did not have, so the emulation would needto do a lot more calculations to perform the same task, and even then it isn't guaranteed that the emulation would work with 100% compatability. Anyway i feel this is getting too far off topic, if you want to discuss this further open up a new thread and we can look into it in detail there.