Tri count

Does anyone know the polycount of the monsters that'll be in doom3?

1500 to 3000 polys I guess. They can't go nuts with the polycount because all the textures and their respective special blending modes applied over them, not to mention how increasibly slower the shadow tracing becomes.

But the polycount of the reference models should also be mentioned! At least they create the detailed look of the bumpmapped surfaces.

Bumpmapping adds detail, but they are still too angular in sections for my liking.

Carmac him self stated that the models wil have polycounts from 3000 til 6000 poly's.. maby even higher

GS-1719 said:

Bumpmapping adds detail, but they are still too angular in sections for my liking.

Well that's the point. Polygons are angular and that can't be resolved until we start doing voxels or start depending on NV30+ hardware.

Tetzlaff said:

But the polycount of the reference models should also be mentioned! At least they create the detailed look of the bumpmapped surfaces.

Why? Even if they're made of 36 billion polygons it all depends on the amount of pixels found in the normal maps. I'm guessing 3 to 4 512x512 maps? Maybe less.

I just said why... the size of the maps is another thing.

Lord FlatHead said:

Well that's the point. Polygons are angular and that can't be resolved until we start doing voxels or start depending on NV30+ hardware.

Voxels is not a good idea...

Displacement mapping would be a very good move from Bumpmapping.

Displacement mapping is downright useless so far. Really friggin slow. It would be wiser to actually model detail and texture microdetail over them.

Zaldron said:

Displacement mapping is downright useless so far. Really friggin slow. It would be wiser to actually model detail and texture microdetail over them.

Yes, but cards like the Parahelia (or however it's spelled) do dynamic tesellation/displacement quite well. It's not so far fetched that post-Doom3 engines could use full displacement mapping over whole scenes.

Voxels, IMHO is really the better direction game programmers should be pushing in. While voxels seem absurdly resource consumptive and have seemingly insurmountable inefficiencies as far as animation and so forth, so did totally dynamic bump-lightmapping when quake1 came out. I often think the industry is focusing so much on making movie-like quality rendering that we forget we're not trying to make movies. Hollow polygonal shells are great if you can (in effect) hand animate something once, and then that's it. But photorealistic games are after the larger goal of simulation rather than (simple) visual quality. Voxels, in the end, should offer better simulation capabilities for the simple reason that they more closely model how real objects are made up.

The problem is that as long as we stack more and more new effects over all the surfaces in the game, a "regular" scene would be too demanding for displacement maps. Polycounts and textures size are ever-increasing. Soon we'll have really high precision, Renderman-style shaders, supersampling, AA, scalable shadow maps and other heavy-duty effects.

It's hard to imagine displacement maps over all of this. The film industry rarely uses this feature, and not many high-profile renderers support it. Sure, technology grows in an exponential rate, but it's hard to imagine game developers taking advantage of this situation, as the gap between old-gen and next-gen becomes bigger.

Voxels are great. We would design our levels using 3D packages, and treat the meshes as volumes for the particle generation. We would need a vid card capable of creating as much voxels as pixels in the screen at really high speeds. Hopefully by then occlusion technology would have evolved so much that all the extra power could be used for other calculations.

The general opinion is that voxels suck ass because these are fixed in the gameworld, but as soon as we can fill volumes with as many particles as we need in order to ensure they all look 1-pixel big (or even less, to archieve supersampling), then we will have fully curved surfaces (as far as you can in a defined resolution).

Zaldron said:

The problem is that as long as we stack more and more new effects over all the surfaces in the game, a "regular" scene would be too demanding for displacement maps. Polycounts and textures size are ever-increasing. Soon we'll have really high precision, Renderman-style shaders, supersampling, AA, scalable shadow maps and other heavy-duty effects.

It's hard to imagine displacement maps over all of this. The film industry rarely uses this feature, and not many high-profile renderers support it. Sure, technology grows in an exponential rate, but it's hard to imagine game developers taking advantage of this situation, as the gap between old-gen and next-gen becomes bigger.

I'm afraid I have to disagree with about all of this ;) Stacking (I'm assuming your talking rendering/lighting) effects is an entirely different issue from using a texture to produce actual geometric detail. It's like saying that we can't have high-poly models because the cards now support more textures per pass-- while the two certainly influce each other to a small extent, they really are two quite different aspects of rendering: geometry, and the rendering thereof. Displacement mapping is just basically a form of mesh compression/generation. Especially now, film studios use geometry generation/compression all over the place. In a bugs life, all those rocks, plants, and gravel were placed using textures-- this is just a fancy displacement mapping of the basic terrain geometry. Actual displacement mapping was used all over the place, too. In modern film, it's impossible to hand model all the details of say, a canyon. The canyon is just a bunch of fancy input bitmaps, and the geometry is produced LOD style, independant of lighting or shading effects.

Carmack actually thinks (or thinks for Doom3, at anyrate) much like you do-- that it's just better to define the complete geometry before hand, and just sending it speeding though the vid-card. I think this is wrong, and you can see in the Doom3 screen shots, when the characters and objects get close, that they seem angular. It's impossible to just define a LOD fine enough (even practically speaking) to effectively work for meadium sight range as well as close-up switch turning range. We see it in every game-- it looks great when it's at the far wall, but when you get close, that switch we see is just a box, and it kills some of the immersion. Displacement mapping would, in some areas, correct this-- the geometry would still be a box, but you could layer more and more geometric detail-- a texture for screws, then a procedural texture for tiny dents and dust particles when you get really close... This would be the right way to do things-- we already do this with detail texture mapping, and we need to extend this to the geometry, also, to get the full and proper effect.

Apropos developers not taking advantage of the tech: of course most of them won't. It doesn't mean it's the wrong thing to do.

Zaldron said: Voxels are great. We would design our levels using 3D packages, and treat the meshes as volumes for the particle generation. [/B]

The input is not important, but basic data-structure that voxels give us-- walls suddenly have stuff in them by default: gravel, or wires, all detail independant. Toxic waste doesn't leak from a barrel because of some programmed effect, but simply a result of one voxel corroding another voxel, and then passing out rhough the hole. Water will finally lap properly at the shore, without having to resort to artist-worked texture effects-- it will just happen properly! The point of voxels is that designing and simulating worlds, not just the APPEARANCE of worlds, will be much more strait forwards and intuitive. I know it doesn't look at all reasonable now, and by the time it comes around it probably won't be called "voxels" any more. But volumtrically descretized objects are the proper way to go about creating a realistic simulated world, I think.

Wouldn't all those voxels, especially if processed individually for the physics, require insane quantities of RAM and CPU power?

Fredrik said:

Wouldn't all those voxels, especially if processed individually for the physics, require insane quantities of RAM and CPU power?

Absolutely. But so would totally dynamic per-pixel bumpmapping, when viewed with the technology of the Doom1 era.

My whole point is somewhat like Carmacks: while it really can't be done with the technology we have now, we should nevertheless be pushing in that direction, rather than in a direction that is ultimately sub-optimal for use in games despite it's immeadiate "wow-factor". We're stuck using polygons, I think, simply because changing direction means almost throwing out all that technology that NVidia and ATI have spent all this time developing, drastically changing the way models are created, drastically changing the way texture artists deal with models (or will they morph into general artists, now?), drastically changing everything. It's just that polygons work so well NOW, and there's still a lot that can be done with them for even more amazing (and hence financially profitable) results. But it still is really a kind of dead end for games, and I'd like to seem companies at least sneaking voxel type stuff in around the corners. We have volume textures now, which is an elegant start, but I'd like to see more optimization there-- it should be looked at as a standard rendering primitive, not an expensive "effect".

Zaldron said:

1500 to 3000 polys I guess. They can't go nuts with the polycount because all the textures and their respective special blending modes applied over them, not to mention how increasibly slower the shadow tracing becomes.

i think 5000. it looks like a cgi movie so far.

Exactly, the problem with voxels is that sometime, somehow, someone's gotta throw all the poly-based hardware features of the board and aim at a completely different focus. This will render the card absurdly useless for 3D games. Or they could throw in support for both types, but nonetheless they'll have to eventually stop optimizing and upgrading the poly aspect and move on to voxel stuff.

Which company will be brave enough to do so? And who's gonna drive this infrastructure change? id?

The creatures they showed so far can be done with 3000 polys. Carmack mentioned something about more complex creatures requiring up to 5000, more likely bosses.

Zaldron said:

Which company will be brave enough to do so? And who's gonna drive this infrastructure change?

Absolutely spot-on points. Sigh...

Now that I realize every single attempt at improving the way computers handle graphics are a result of the film's industry advantage on this aspect. All the hot features of today's games where taken for granted years and years ago in 3d packages. Particle interaction will be pretty for some odd, impossible to draw right effects, but I don't think the film industry is going to take much advantage of voxels. They don't quite need it.

So who's gonna develop this technology first? It has very practical uses concerning the simulation of a broad range of natural phenomena, but somehow I don't see the Pentagon or some Tokio-based laboratory equipment can influence the technology normal people uses for entertainment.