When I replied to the other post, the implication that PhysX as a physics API is exclusive, is wrong. PhysX can be used by everyone and they do not need to have a NVIDIA card to use it. PhysX calls sent directly to a GPU (GPU physics) are however tied to NVIDIA hardware, since no one else can do it at this point (not because NVIDIA blocked them, they just didn't license the tech). If you don't have a NVIDIA GPU, then obviously GPU physics won't work with PhysX and the physics processing will be done on whatever is the default hardware component (typically the CPU).
And in the case of this game, physics effects that would work perfectly fine on most CPUs was outright removed. Not made into a physics option slider for those with more powerful CPUs to take advantage of, but straight up removed. THAT is bullshit. THAT is exclusive behavior. PhysX itself isn't, but what this game did IS.
As for what you linked to, it's pretty much the standard practice and I'll explain in a very simple manner. Having one video card, render graphics and process physics is "simple". There is no context change and the data is present in the same GPU that can act/react to what was calculated in both graphics and physics.
Having different video cards, one for graphics and another for physics is quite another thing. Even more so when they are from different brands. NVIDIA can't guarantee proper functionality with cards from other manufacturers, if they (NVIDIA cards) are only being used for physics.
People that complain either don't know how it works, or think that NVIDIA is in the business of charity or something and that they need to invest in a tech and still support other manufacturers that will reap the benefits of that tech, without effort. If other manufacturers want support, then they have to work with NVIDIA so that they get support.
That is a bullshit argument and you know it. Using two nvidia cards (1 for graphics, 1 for physics) vs. 1 ati card for graphics and 1 nvidia card for physics is the same damn process. The game makes graphics calls which go to the driver and to the card. The game makes PhysX calls which go to the driver which go to the card. The two are entirely independent flows that are in no way linked. At all. The nvidia card rendering physics operates completely independently from the nvidia card rendering graphics. One the same card they sort of are, yes, but when there are two physical cards they aren't. Also, when using a single card there *IS* still a context change. The process if *MORE* complex for a single card. The graphics rendering also cannot directly access the physics results like you imply - it doesn't work like that. And afaik CUDA still works with an ATI card doing graphics. Since PhysX runs on CUDA, there isn't a single technical reason for this limitation. None. It is entirely a marketing move by Nvidia, nothing more.