the orientation of the future engines
by Paolo Oliverio · in General Discussion · 05/11/2003 (4:41 am) · 5 replies
what will be the orientation of 3d engnes on pc in your opinion?
WE will see others polygon troughput limitative bsp engine for interiors in future?
what's the future precalculated radiosity,per pixel lighting and shadows or per vertex with a lot of poly?
normal map from hi-poly or true real hi poly whith patch or nurbs tesselated in hardware.
there will be a future for the displacement mapping of directx9 hardware(like latest matrox gpu)?
Roam terrais or brute force?
what about procedural texturing(wood shader for example)
write here your opinions.
WE will see others polygon troughput limitative bsp engine for interiors in future?
what's the future precalculated radiosity,per pixel lighting and shadows or per vertex with a lot of poly?
normal map from hi-poly or true real hi poly whith patch or nurbs tesselated in hardware.
there will be a future for the displacement mapping of directx9 hardware(like latest matrox gpu)?
Roam terrais or brute force?
what about procedural texturing(wood shader for example)
write here your opinions.
#2
quickly navigate to the nearest bedroom to play on a computer ;o)
05/11/2003 (11:47 am)
The long term future will have games so realistic and true to life that gamers willquickly navigate to the nearest bedroom to play on a computer ;o)
#3
what spherical harmonics and liniear wavelet lighting are?a tecnique based on lookup maps?please give me a link to this argument.
05/11/2003 (12:17 pm)
few days ago i find a very interesting demo with src on nvidia's site that show displacement mapping also on directx8 compliant hardware and most important with realtime level of detail but like Sander say is a bruteforce terrain tecnique looking the wireframe.xvox,this is its name,have a good poligon troughput.i don't have read the source but i see that this demo isn't limited by fill rate or polygons(adjusting the level of detail).I must understand the bottleneck of this tecnique.cpu and locks?i have to read the source.what spherical harmonics and liniear wavelet lighting are?a tecnique based on lookup maps?please give me a link to this argument.
#4
look at the series of other tecniques birth after the cell shading.
do we really want photorealistic look in all games generes?
for example i don't think that you will like a photorealistic platform game.
05/11/2003 (12:33 pm)
i think that the future in term of game graphic will not point to the realism that will be the omologation of game graphic but on surrealism,strange graphic styles and on the improvment of non-photorealistic tecniques.look at the series of other tecniques birth after the cell shading.
do we really want photorealistic look in all games generes?
for example i don't think that you will like a photorealistic platform game.
#5
As for surreal rendering.. it will only be surreal if it borders on real.
As for the fill rate limits, they'll be hit hard by the techniques used in the next generation of engines like doom3, halo2 etc.
As we start using pixel shaders more and more, and we store and retrieve more data into and out from the framebuffer using all kinds of techniques like pbuffers etc. fillrate will become our no. 1 enemy.
Just look at doom3, the models aren't really high poly, they use bump mapping to make the models look more detailed.
They have to be low poly because of stencil buffers, which is rather cpu limited.
But even before the cpu limit is hit, the fillrate limit is hit long before that.
Stencil shadows draw huge screensized polygons several times in the stencil buffers.
Other techniques like shadow projection are more efficient when it comes to more complex scenes (compared to stencil shadows) but they require that the same scene is rendered at least once per light + 1 layer per light for the scene from the viewpoint of the camera
(unless you use deffered rendering, in which case you need to draw a screensized polygon per light)
edit: as for spherical harmonics, look at the following link here
05/12/2003 (4:41 am)
Good point, there will also be a lot of development in non-photo realistic rendering.As for surreal rendering.. it will only be surreal if it borders on real.
As for the fill rate limits, they'll be hit hard by the techniques used in the next generation of engines like doom3, halo2 etc.
As we start using pixel shaders more and more, and we store and retrieve more data into and out from the framebuffer using all kinds of techniques like pbuffers etc. fillrate will become our no. 1 enemy.
Just look at doom3, the models aren't really high poly, they use bump mapping to make the models look more detailed.
They have to be low poly because of stencil buffers, which is rather cpu limited.
But even before the cpu limit is hit, the fillrate limit is hit long before that.
Stencil shadows draw huge screensized polygons several times in the stencil buffers.
Other techniques like shadow projection are more efficient when it comes to more complex scenes (compared to stencil shadows) but they require that the same scene is rendered at least once per light + 1 layer per light for the scene from the viewpoint of the camera
(unless you use deffered rendering, in which case you need to draw a screensized polygon per light)
edit: as for spherical harmonics, look at the following link here
Torque Owner Sander Van Rossen
Which means that bsp's, at least the traditional ones, will become a thing of the past.
More likely we'll see more portal based rendering, where every portal cell is send to the GPU as a whole (sorted by texture)
as for precalculated radiosity, this will most likely be replaced by spherical harmonics and non-liniear wavelet lighting, which can create simular quality graphics, is also pre-calculated, but as long as the meshes are rigid (ie. don't change shape) they can actually move and have moving lights around them (and the lighting changes accordingly; and as long as the lights don't get too close)
per pixel lighting + shadows is definitely the future, and no engine without it will be regarded as 'advanced'
Displacement mapping is very interesting technology, i'm not sure how it will develop.
The upside is that it can render terrain really quickly and efficiently, the downside is that because it's brute force it eats a lot of fillrate, which is already becoming the biggest bottleneck now that per-pixel shaders are becoming commonplace.
As for procedural textures, in my opinion they've never been as good a thing as some people claim they are.
I've never seen any procedural textures that where better than a good texture.
At most, imho, they're good for noise and adding dirt to images.
I doubt they'll ever be usefull for anything else (but then again, who knows)
edit: i said liniear wavelets, but ment non-liniear wavelets..