correct me if I’m wrong:
current engines uses one transformation matrix for translate/rotate/scale and then data is feed into hardware for rendering - there is another matrix for project 3D vertexes into 2D screen coords right?
well… I got a simple method which uses less compute power and has as a bonus simple distort effects
my method uses 14 multiplications and 20 additions and the classic way uses 32 multiplications and 24 additions
you have some low-level engineer to confirm that there are two matrix? and probably have to talk to nVidia also about this…