http://www.youtube.com/watch?v=9F6zSgtRnkE

amazing.

Yes, tesselation is may really going to be something like bump mapping was years ago...

I guess it's done via geometry shaders? Or is it part of a fixed pipeline?

Yup, done via the new geometry shader units. I'm asking myself though where the border is in terms of actually implementing this in a game. The amounts of work are just incredible when thinking of having to do high-poly versions of everything in your game so you can apply tesselation to the low-poly versions. In order to release games in the current time frames you'd need to hire quite some additional artists to get the work done...

The results are cool though while in that demo things are a bit overused and the performance isn't that great so it's going to be a while until we'll see that level of detail in an actual game...

Look nice, but as said. With all these graphical options, the
raw amount of just creating Model-content that actually fill a
game must be increddible.

I suppose in the future models of all types will be more and more
procedurally created, kind of the Poser/Speedtree way, also
for buildings etc, simply to be able to create it in a reasonable time.

I guess its not the models which have to be more high-resolution but there's some kind of displacement map for the textures. For complex models like the dragon this displacement map is created from a high-res model, of course. But for textures like the stones or the walls I think they tile it, otherwise it would be quite inefficient to create content.

What I don't get, though, is why this should be DirectX 11 only - aren't geometry shaders already present in DirectX 10?

Blisters. That's ingenious!

Of course things will be tileable for repeatable patterns but just unlike for normal maps you cannot fake a good final look by just running some filters over the texture itself. You have to create it from a high-res mesh as it usually also gets done for normal maps (I don't think handpainting the needed maps is an idea that works). So as final thought you'd need a high-poly mesh for every single texture you paint...


They used to be in DirectX 10 but NVIDIA refused to implement them back then and so Microsoft agreed in delaying them. That also lead to ATI having next to everything since the HD2900 GPUs which might have been one of the reasons why those cards turned out so poor as lots of time was invested into integrating a "useless" element on the chip...

That also might have added as to why DirectX 10 did get so little attention as it in the end only delivered little improvements or new features. It's now DirectX 11 which introduces all of this and gives GPUs the possibilities to on-the-fly create such geometries. I'm not sure when this will make it into OpenGL - considering the failures of version 3 it probably will take a while...

That looks just awesome.

Now, I don't get the fuss about the increased workload -- I disagree that a good displacement map cannot be hand-drawn. It can. It requires no more fidelity than a height-map used for parallax occlusion mapping.

The stone walls were almost definitely hand-modelled, but I don't see a reason for the rocky-road to be.

Also, not every surface requires a displacement map -- those areas covered with grass would just waste resources if they had a fancy displacement map; other things that the player isn't meant to get close to won't need to tessellate at all. For many surfaces a simple normal map will suffice -- this demo was intentionally very bumpy.

But I very much look forward to using DirectX 11 -- at first only DirectCompute and Direct2D caught my fancy, but now the tessellation is really enticing That dragon looked SO cool!

Jibb

amazing