Well... GPU physics is a concept which can use either one graphics card, two graphics cards, or three (2 for graphics and one for physics). In either setup, the graphics card you set as doing the physics calculations doesn't need to be the same card as the one doing the graphics. This means you can throw in a cheaper card for physics work, whilst the more expensive one(s) do(es) the graphical work.
Driver wise, this sort of configuration needs it own physics API, something like the DirectX standard for graphics, so interoperability can be accomplished between different solutions. Then you need the hardware drivers, the necessary hardware to support the configuration, and then most important of all, software support. Until Developers start taking GPU physics seriously, it ain't gonna come round.