First theoretical question. What is better (faster)? Develop your own gpgpu techniques for physics simulation (cloth, fluids, colisions...) or to use PhysX?
(If i say develop i mean implement existing algorithms like navier-strokes...)
I don't care about what will take more time to develop. What will be faster for end user? As i understand that physx are accelerated through PPU units in gpu, does it mean that physical simulation can run in paralel with rastarization? Are PPUs different units than unified shader units used as vertex/geometry/pixel/gpgpu shader units?
And little non-theoretical question: Is physx able to do sofisticated simulation equal to lets say Autodesk's Maya fluid solver?
Are there any c++ gpu accelerated physics frameworks to try? (I am interested in both physx and gpgpu, commercial engines are ok too).
Edit: Simulation does not have to be realtime i'm thinking just about acceleration.