DX11 generation versus DX10

[cybereality]

I think this can be good in many cases and should help improve the graphics of games. If anyone is familiar with 3D modeling packages it is very common to use this effect for pre-rendered content (mesh-smooth in 3Dmax or Sub-division modeling in Maya). In the majority of cases this can add extra geometry without affecting the general topography or normals so it retains the same shape (just at the higher mesh density).

That air canister is just a simple example, I think some of you guys are missing the point. Yes the devs could have just made it a high-res mesh. But then performance would be lower. In order to compensate they will use dynamic LOD (level of detail) which means the artists have to go in there are output multiple versions of each mesh (lets say 3 or 4 versions at various detail levels). This is more work for the artists and does not look so great in games. You know when you are playing a game and looking at say a rocky mountain and then as you move closer the geometry just snaps/pops to a different formation? That is LOD at work. Dx11 tessellation fixes that. So ideally you would get a higher mesh density (ie more poloygons), better performance, and smoother transitions. Its a win win situation assuming it is implemented correctly in Dx11 (which is yet to be seen). There also has to be a way to customize the tessellation so games will need to be specifically built for this feature. But it has a lot of potential.

FYI: I think the first game to use this technique was Messiah which came out like 10 years ago. I would be interested to load that game up today and see if it actually scales...

No so bad for 1998.

 

[vjcsmoke]

I would too if there was a choice.(I will buy a next gen DX11 card) The 295 does have support for DX11 features and as far as I understood, some of the effects can be downscaled on DX10/10.1:

No it doesn't. The GTX 295 is a DX10 card. The chart of DX11 features you quoted is for the Radeon 5870.

Looks like prices of the 295 have taken a small dip.

EVGA Co-op edition is $456 after bing cashback.

http://www.bing.com/shopping/eVGA-G...55996E4CDE461DE6F857"]&wf=Commerce&FORM=ENCA2

 

[rincewind]

I think this can be good in many cases and should help improve the graphics of games. If anyone is familiar with 3D modeling packages it is very common to use this effect for pre-rendered content (mesh-smooth in 3Dmax or Sub-division modeling in Maya). In the majority of cases this can add extra geometry without affecting the general topography or normals so it retains the same shape (just at the higher mesh density).

That air canister is just a simple example, I think some of you guys are missing the point. Yes the devs could have just made it a high-res mesh. But then performance would be lower. In order to compensate they will use dynamic LOD (level of detail) which means the artists have to go in there are output multiple versions of each mesh (lets say 3 or 4 versions at various detail levels). This is more work for the artists and does not look so great in games. You know when you are playing a game and looking at say a rocky mountain and then as you move closer the geometry just snaps/pops to a different formation? That is LOD at work. Dx11 tessellation fixes that. So ideally you would get a higher mesh density (ie more poloygons), better performance, and smoother transitions. Its a win win situation assuming it is implemented correctly in Dx11 (which is yet to be seen). There also has to be a way to customize the tessellation so games will need to be specifically built for this feature. But it has a lot of potential.

People would be less skeptical if they'd ever done any modelling with subdivision surfaces, I'm sure. The biggest problem with high-polygon models is animation, be it keyframe or skeletal, as you have a lot more vertices to transform. Subdivision circumvents this problem as you are operating on the original geometry. The memory footprint associated with these additional high-detail models is another issue that subdivision ignores. I'm not sure that content creation time is a fair point though, as what we're talking about would be as simple as performing N levels of subdivsion and saving that as a new mesh (any vertex tweaking after that is adding details you wouldn't receive from tesselation).
The need for more than one LOD model won't be eradicated in games with large view distances. Tesselation doesn't work well if the control mesh is too primitive, so you'd need a base model suitable for this purpose and extra low-poly versions for objects a long distance from the camera.
'Popping' is actually relatively easy to workaround. The higher LOD initially shares the same exact topology but with more vertices, and as you move closer it is morphed to flesh out all the details. Unfortunately, this is a costly process. There are 'continuous LOD' terrain engines which use a similar technique on height-map data.