1.
"You said CPU and GPU are totally different"
Yes, coding for CPU and coding for GPU is different, they work in entirely different matter. You can't just offload "whatever the heck you want". GPUs have the ability to be amazingly fast, while CPUs are, as of yet, much more precise than GPUs. Try Googling Mental Ray, Vray, Arnold (CPU renders) and FurryBall (GPU renderer) and see how much GPU rendering sucks compared to CPU rendering. RealFlow or any other types of liquid simulations. (hint - 99% they run on CPU). nVidia is trying to push their CUDA for physics, but CPUs are, and at least up untill 202 or so, will be far more accurate.
2.
"and you cannot offload work of CPU to GPU"
No, I said there is "GPU acceleration" in my 1st or 2nd post. But 90% of the software does not use it. The 10% of software that does use it - well not to a great extend. One of the best and most noticeable GPU accelerations are in Sony Vegas (OpenCL). Still, only a minority of functions is accelerated on a GPU. And in games - the CPU and GPU act as how the code was written. If the simulation is to be calculated on the CPU - then CPU it is. If it is send to the GPU - then GPU it is - and there is nothing none of us can do to change that.
3.
PhysX is a minority of usages. Get it, minority. There are so many other engines and almost all of them run on CPU. And Alan Wake is a game that does not have PhysX. The Havok runs on CPU.
4.
As I said before, drop this DX 11 update / DX 12. It is not here. It won't be here for a while. Whatever it promises, it does not help the person who published this thread. He has a question now, not when those updates come.