so... im watching the DX11 tech demos and every single one of them is hyping the tesselation feature:

like here:
https://www.youtube.com/v/bkKtY2G3FbU

or here:


it actually is not new... i remember seeing it aplied on some tech demos back on DX9 SDK

but what im wondering is: if the purpose of game modelers engaging with normalmaps is to reduce the monsterous polycounts... why take all that trouble just to get monsterous polycounts again back on the graphic card when the algorythm just tessellates your model as he likes?

does it means game modellers will have their poly budgets expanded to 1 or 2 million polys?

why not take the model directly from zbrush/mudbox instead of taking all the trouble to create a low poly mesh with normalmaps that will just get tessellated to a symilar polycount as the original model?