Radiosivity question

Is very hard to implement a basic radiosivity in hardware?
If the computational work per triangle is very hard the meshes could be aproximated with NURB surfaces, for example, and these surfaces stored in video memory where they can be used to compute the radiosivity.
And i mean to store the surfaces besides meshes triangles (all the surfaces of the scene should stay in memory, but only the visible meshes should be drawn).
Finally, when the card is doing per triangle setup, it can take an intensity level calculated from the NURB surface and modulate it with per-pixel fragment.
If my logic is wrong please tell me where.

The problem is that radiosity (just like raytracing) is a global algortihm, i.e. it takes the entire scene into account. Thus the hardware would need to have access to the entire scene. You could probably accelerate parts of the process though, if you settle for a hemicube approach for example, it’s certainly possible to do visibiltiy calculations with a hardware z-buffer.