GPU Occlusion Culling Only?

Hello there,

I'm looking for a library with approriate API for running a very efficient Occlusion Culling using a GPU and/or multicore CPU.
I do NOT want to do the rendering stage.
Formally, I need to do the following:

Input:
1. list of triangles, representing a 3D object surface
2. 3D view point
output: subset of triangles which are visible from the given view point.

It should handle a 10-20M triangles list and I need to make it for about 100K view points (possibly in parallel).
Any suggestions?

Thanks a lot,
--Amir