It's a fair point, the techniques used to tessellate do offer diminishing returns, as does any iterative process.
The problem here is that for AMD to state it, shows that they think their rivals perhaps offer better and faster solutions than they can and do.
it should go without saying that any techniques available to devs that intrinsically increase a scenes complexity HAS to reduce the speed at which hardware can render it. The tessellation seen in DX11 will be extrapolating the extra detail based on a less detailed model. Where this is great for me is in scalability, you can suddenly do more than just designing models of different complexities and offering a fixed switch in the menu, you can now alter the complexity of your scene on the fly. However as with any tunable parameters, you have to be sensible with it!
I recently went to a conference where AMD, Intel and NVIDIA (amongst others) presented, just general overviews of their up and coming technology and no matter what AMD did, whether it was re: OpenCL or their CPU road-map, they somehow came across as a bit desperate. I mean their processors actually do very well in the supercomputing world, yet they gave a presentation BEFORE Intel which felt nothing more than a catch-up exercise where their slightly stammering presenter convinced us they were competitive...
They really need to stop talking and start doing again, they have everything at their finger tips and are a strong company yet always seem to come across as slightly desperate underdogs.
)
