in a nutshell, nvidia has made some significant changes to the cuda framework, all of them aimed at making it easier for a programmer to exploit the power of a gpu.

of course, there are few programmers outside of specialized scientific and business areas that have the skill or desire to exploit the power within a general purpose capabilities of a gpu, so for us in the desktop world it will probably amount to a hill of beans, but it's still nice to see nvidia working on refining cuda.