Not sure if it'll comply with 10.1, since 10.1 demands control over AA sample positions by the application. NVIDIA claim fixed non-programmable grids again for pixel subsamples in G80. It seems to comply otherwise, although 10.1 isn't finished so nobody knows and WDDM 2.1 requirements might scupper hardware outside of the API, depending on what's tied to what.
As far as precision goes in 10.1, probably best not to confuse "precision improvements" as a jump to double precision floats. FP64 isn't happening (IMO), but there'll likely be a tightening to 754 where it deviates in places in 10.0 (blending and filtering aren't specced to round to nearest even in 10.0, just 1ulp, f.e.).