The first DirectX 11 benchmark

[Gloglebag]

http://unigine.com/press-releases/091001-directx11/

http://unigine.com/press-releases/09102 ... benchmark/

[Section_Ei8ht]

Damn... I remember following the Unigine back in high school when it was open source. That little engine's come a looooong way.

[cincinnati]

why must these advances be at the expense of my feet on that footpath?

[Lord Ned]

I downloaded the benchmark for kicks... I'm impressed. They've nailed the perfect settings for day/night and I'm needing new pants after scrolling back and forth on the time of day bar.

Too bad I don't get the tessilization on my Dx10 card.

[Nysuatro]

This is amazing. Can't wait to work with this new tech.

[dux]

Well I have a 5870 but I run xp 64 so cant use this in anything other than OpenGL lol

[Minos]

I wonder what impact automatic tessellation will have in our workflows. Do the assets need to be prepared in special ways to get it to work? Anyone here familiar with it could give us more info?

[Lord Ned]

I wonder what impact automatic tessellation will have in our workflows. Do the assets need to be prepared in special ways to get it to work? Anyone here familiar with it could give us more info?

I'd assume it would require some sort of higher-poly mesh for some things. If you looked at the Dragon, when they turned the tessellation on his spines grew out and curved... Where would it get the data for that (+the curve)?

[Zacker]

Potentially it could get most of the data from the normal map?

[Furyo]

Yeah the only way this is a major improvement is if the entire data is stored in the normal map and it's just procedurally generated. If you still have to model all of this, the only improvement is "look we can now render even more polys" and we've been there done that...

[Warby]

yeah i am very puzzled where the engine gets all the extra 3d information from if its up to the artist to build the super highres stuff and really this is just an interpolated lod method ... than this would get a huge fat *LAME* label from me !

[Lord Ned]

Potentially it could get most of the data from the normal map?

You could get a protruding horn but it would protude based on the surface normals... There'd be no way to make it curve.

However: I could see the tessellation being done via normal maps on the rocks and the ground and the window ledges and stuff, because it's straight out. I think it's basically taking the normal map data and turning it into real 3d geometry, which keeps the visible 3d effect even when you get next to it which normal maps do not.

I dunno, will be interesting to see where this goes.

[Seldoon182]

When the tessellation will be available with CryEngine [HP]!?

[-HP-]

When the tessellation will be available with CryEngine [HP]!?

I don't know, but I would sure love to play with that beauty on CE! Tessellation might revolutionize the way we make games, like normal maps did a few years ago.

For those asking how this works, I did a few tests with a buddy of mine a few weeks ago, and It's pretty easy. It works pretty much like Parallax mapping, but tessellation creates geometry.

You got to be careful when modelling high polies for this, and avoid 90º angles, otherwise you will have some artifacts. We can still use floaters on the high poly like we did for normal mapping, but we gotta be careful not to exaggerate, and the angles are very important. And then you would bake the object as you always did, normal map, AO, etc, but now you also need to bake the height map, which is basicaly a black and white map, that goes into the alpha map of the normal map.

Here's an example of one of the textures i did a few months ago.

So yeah, as you can see if you have a good high poly, is pretty much just another map to bake.

You can also extract a heightmap from a normal map generated from crazybump for example, to use on terrain or those rock walls shown on that vid, but the artist got to be very careful with the artfcats, and try to get a very smooth heightmap and avoid abrupt transitions.

[Buddy]

It's like relief mapping but it results in actuall geometry (and it can 'go out' of the surface, so the volume/deepness is "real"), propably uses normal+height.

I don't see a revolution anyway, a lot of artifacts here and there.