You cant divide 256 by 48 maybe 240 but i realy doubt it will have more than 192 sps.
And who said it will be based on 48 SP wide SMs? It could have 8 SM with 32 SP each, divided into 2 GPCs, that is, exactly half of a GF100/110. This way it would retain the same setup and geometry (+tesselation) power as the GTX460, maybe a wise thing considering that performance claims put this card in the same performance level as the GTX460 and to mantain a minimum geometry capability which could come in handy for future DX11 games. Basically this could be a way to create some predefined levels of geometry power available to game developers*: one high-end (GTX 480/580 etc) and one mid-range (GTX460, 550, 560, etc). AMD already did this, more or less, although indirectly, when they designed their architecture. Basically every AMD card has the same geometry and tesselation unit/pipeline, a single one, except Cayman which has 2.
* This way they could create 2/3 levels of geometry complexity - independent from shaders/filters etc- instead of having it be based on a linear progression. What I mean is that since Nvidia cards' geometry/tesselation power raises along with the number of SMs, Nvidia may try to create groups of similar geometry capabilities in order to make developer's life easier. i.e when it comes to geometry the GTX 550, 460 and 560 would be able to use the exact same complexity, with small natural variations that would benefit the faster cards, and since for example the GTX560 has far more shaders, you would be able to raise shader complexity, filter quality etc. But at any rate the geometry complexity is assured to be consistent for every segment in the lineup.
I'm very bad explaining myself. I just wrote the same thing 3 times over in this post (a bit differently each time) and I'm still not sure if it's clear... lol i hope so.