See the A7 demo in user contrib?

I got into a discussion there and thought best to move here.

I was hoping someone could give me an overview of what happens inside GS.

Given a shader/entity situation:

Vertexes go through a series of transformations ending up in screen space, then data is passed to the shader and at this point and the hardware takes over here, or?

As far as culling, this is what the BSP tree is for? But I remember reading that this scheme is outdated? This doesn't help entity rendering does it? What is next? How important is it on modern machines if you are implemnenting good LOD?

What percentage of the fps hit of putting a polygon on the screen is done by the hardware and how much the engine? What other issues are at play?

GS is great software - Content, art, sound, music and mainly 'idea' make a game - And with specular, normal, relief mapping, bloom shaders, real time mirror reflection, physics, and a strong community... well, I am here because I believe that there is enough here to make a viable current gen game.

(blended vertex weights and decals not withstanding)

Still curiuos though. In terms of culled, shaded polys rendered, how does GS stack up to modern angines? A7?

Help or links always appreciated.