I was wondering about the whole 9400M in iMac thingie. People are asking themselves where's the benefit of integrated over discrete and frankly, it might be hard to see. But one thing occurred to me; the integrated GPU shares the RAM and that can make a difference. While you'd think that it's rather a bad thing (hey, stop stealing my RAM! got get your own you b#(*!), I think it might be for the better. How? GPU accelerated computing. I've been playing with CUDA lately and however OpenCL turns out to be, the biggest problem will be essentially the same - moving data from RAM to GPU memory. This takes time. Moreover, you have to do it twice! Say, you want to multiply two numbers on the GPU. Initially they're somewhere in your main memory. You have to dig them up, transfer them to GPU, multiply and transfer them back again. And graphic cards aren't really designed to do that - normally data comes through one end, and goes out through the other (to the display!). If your GPU could simply access this data directly (I think you'd have to do some hardware/bios/software magic to do that), it would make things much nicer ;-D I need to look it up. It can be quite interesting. Anyone did some CUDA programming?