But why rotate the whole world around instead of rotating the camera? any benefits to this?
Are we always going to maintain a fixed camera? Why would we apply world transformations instead of applying transformations to the camera?
Can we use different perspective projections to simulate different types of lenses? Or do we usually do that in post processing?
In general, which stage of this pipeline would be considered to be the bottleneck? I would assume it is the part where the transformations are applied, as that would take the most arithmetic instructions.