The impact of streaming architectures on Voxels?

[MrWibble]

when did i say this? i can't remember saying (or typing, as the case may be) this, nor do i believe it. LOD will always be an issue.

You said: "the problem with voxels will always be that they look poor at low resolutions. the problem with polygons is the opposite, because at higher resolution you can start to see the geometry that makes up an object. neither solution is artifact free."

According to this statement, polygons are bad at higher resolutions and voxels are bad at low resolutions. This makes no sense to me. If you fix the number of primitives then voxels will look always worse at higher resolutions whereas polygons will only look bad if they approximating a curved surface, and even then it is only a minor problem. I'm not sure why either approach would look bad at a low resolution.

This is why I said it's simply a matter of LOD - the quality depends on the ration of how many primitives are used to draw things that require a lot of pixels (either being close to the camera or running at higher res) versus how many are required to accurately model a surface.

Earlier on you said that if performance wouldn't be a problem voxels would be limited only by output resolution and things could be really organic looking. Again, to me your statement suggested that in some way voxels would actually be suited to this job, other than an issue of performance. Also there is an implication here that polygons or more traditional approachs in some way would not be good at this job, even on your system of ultimate performance.

That's why I keep poking - every time you agree in principal you throw in a comment about voxels that makes me scratch my head and wonder what it is that you think voxels can do that I don't.

 

[purpledog]

That's no different to how fur and grass is rendered now (except typically it would use a 2D texture and not a 3D one). The example was using voxels to provide a fully deformable pitch, not make some grass wave about.

Ok, if you want to represent the entire world by ONE volume (even compressed by an octree), you won't go far. But grass and fur are a bit more subtle here using "approximative instanciation".
so the idea is: yes, some stuff are literally impossible now, but there's a lot of more subtle methods around. I'm sure we'll see volumetric trees, smoke, and other funky stuff with next-gen console.

Just don't bury an idea because somebody is far too much enthusiastic saying:
"just one octree node (and subnodes of the same type) representing everything. "

 

[Rodéric]

Outcast (1999 PC game) had a nice Voxel tiled heightmaps based Terrain engine making look very life like, but it suffered from a little drawback, you couldn't have true 6DOF since at some angles the terrain would look rather... awkward...

Voxels can be used today in a limited set of circumstances, and they do the job just fine. (Commanche, Delta Force also featured Voxel Terrains)

 

[MrWibble]

Just don't bury an idea because somebody is far too much enthusiastic saying:
"just one octree node (and subnodes of the same type) representing everything. "

I don't think I did bury the idea - I posted the 3rd post in this thread in which I clearly said that I thought volumetric techniques would work for some particular effects but not be a general solution (i.e. exactly what you just said...)

However a few people have made what I consider to be odd and baseless claims about what voxels can do and what advantages they might have over a traditional polygonal appraoch, and I think it's reasonable to challenge those.

 

[Shifty Geezer]

Yes. Please challenge it. Please discuss and dissect voxel technology to it's fullest. It's the oly graphical technology debate on this forum at the moment - we've got to milk it for all it's worth!

 

[MrWibble]

Outcast (1999 PC game) had a nice Voxel tiled heightmaps based Terrain engine making look very life like, but it suffered from a little drawback, you couldn't have true 6DOF since at some angles the terrain would look rather... awkward...

Voxels can be used today in a limited set of circumstances, and they do the job just fine. (Commanche, Delta Force also featured Voxel Terrains)

Odd definition of "today". Last games from Novalogic that I saw had moved over to using polygons - and were looking much better for it.

At the point where software was still keeping up with hardware rendering, or where hardware was not available, voxels were a viable alternative technique to writing a polygon rasteriser. Looking back at screenshots of those titles now and the artefacts are only too visible even at the low resolutions they were rendered at.

I don't think I'd use a voxel technique for that kind of geometry today. Is there anything actually still using them for anything solid and/or large scale?

 

[Platon]

Outcast (1999 PC game) had a nice Voxel tiled heightmaps based Terrain engine making look very life like, but it suffered from a little drawback, you couldn't have true 6DOF since at some angles the terrain would look rather... awkward...

Voxels can be used today in a limited set of circumstances, and they do the job just fine. (Commanche, Delta Force also featured Voxel Terrains)

Outcast... mmm... my favorit game so far. Yep, their voxel based enviroments rocked, and even to this day, I admire how "organic" they look, I guess mostly due to the non flat appearence that the polys are associated with...

 

[inefficient]

Outcast... mmm... my favorit game so far. Yep, their voxel based enviroments rocked, and even to this day, I admire how "organic" they look, I guess mostly due to the non flat appearence that the polys are associated with...

I don't know abou the to this day part. For it's time it was pretty refreshing.

The strong reason they chose voxels was because they intended to release a game that did not requite a 3d Accelerator. Which at the time was still not mainstream. Same with Novalogic's Delta Force and especially Comanche.

If you compared those SW rendered voxel games agains other SW rendered games which did not even have bilinear filtering you would be on fairly even ground.

 

[Shifty Geezer]

Wonder what state Voxels would be at if GPU designers went with that strategy instead of polygons...

 

[Platon]

Wonder what state Voxels would be at if GPU designers went with that strategy instead of polygons...

I wonder that my self, hardware accelarated voxels, hmm...

 

[Platon]

I don't know abou the to this day part.

I guess that was a minimal overexaggeration of my part ...

 

[MrWibble]

Wonder what state Voxels would be at if GPU designers went with that strategy instead of polygons...

I suspect they wouldn't last long in the mainstream graphics board industry. Medical imaging or other areas where solid model visualisation is important probably already have a market for such technologies so it's not like the tech is forgotten - it's just not seen as particularly worthwhile for this arena.

Even ignoring performance they do not adapt well to many situations. Certainly memory would be an extremely serious issue for a consumer product.

Also even the games that used voxels did not use them exclusively - Outcast had polygonal characters for example. Trying to do character animation using voxels would, I imagine, be a bit of a nightmare. I still haven't seen anyone suggest much of a solution to the issue of shading them either.

In short, although some kind of acceleration would probably give you the ability to make a faster and prettier game than the software-rasterised generation of voxel games, I think they'd be noticeably poorer than the polygonal approaches everyone has adopted.

Of course as traditional 3D boards get better at doing general purpose operations (the introduction of shaders and especially flow control) but still with very fast operations for sampling data, they are approaching the kind of hardware you might want for accelerating voxels anyway.

So I think for the areas where a volumetric approach works, they will happen with the hardware we already have (or are about to get). I doubt we'll be regretting not having voxel chips.

 

[see colon]

You said: "the problem with voxels will always be that they look poor at low resolutions. the problem with polygons is the opposite, because at higher resolution you can start to see the geometry that makes up an object. neither solution is artifact free."

According to this statement, polygons are bad at higher resolutions and voxels are bad at low resolutions. This makes no sense to me. If you fix the number of primitives then voxels will look always worse at higher resolutions whereas polygons will only look bad if they approximating a curved surface, and even then it is only a minor problem. I'm not sure why either approach would look bad at a low resolution.

i already clarified this, but i guess you missed it. i was expanding on this statement..

if voxel rendering performace wasn't an issue, voxels would be more suitable because things can take a really organic shape, limited only by output resolution.

in the not so distant future, however, it's not going to be a problem, since we are reaching a point where we can render more polygons than we have pixels.

let me rephrase that so maybe you can understand it better.

the only way voxels will be more suitable for phot-realistic rendering is if performance and resolution are infinite. pretty soon we'll have hardware that can render so may polygons it won't matter, because each pixel will be occupied by 1 or more polygons..

This is why I said it's simply a matter of LOD - the quality depends on the ration of how many primitives are used to draw things that require a lot of pixels (either being close to the camera or running at higher res) versus how many are required to accurately model a surface.

Earlier on you said that if performance wouldn't be a problem voxels would be limited only by output resolution and things could be really organic looking. Again, to me your statement suggested that in some way voxels would actually be suited to this job, other than an issue of performance. Also there is an implication here that polygons or more traditional approachs in some way would not be good at this job, even on your system of ultimate performance.

no, you're putting words into my mouth. what i was saying is that you'd need an infinite amount of performance to get close to photo realism in realtime using voxels.

 

[amk]

Voxel terrain tutorial:
http://www.flipcode.com/articles/voxelland_part01.shtml

Novalogic's Delta Force 3(?) achieved 6DOF. Rather than drawing parallel lines, it drew lines to a point. I found a discussion of it years ago on the forums of an engine called TerraVox, which seems to have vanished from the Web.

 

[MrWibble]

i already clarified this, but i guess you missed it. i was expanding on this statement..

let me rephrase that so maybe you can understand it better.

the only way voxels will be more suitable for phot-realistic rendering is if performance and resolution are infinite. pretty soon we'll have hardware that can render so may polygons it won't matter, because each pixel will be occupied by 1 or more polygons..


no, you're putting words into my mouth. what i was saying is that you'd need an infinite amount of performance to get close to photo realism in realtime using voxels.

Phrased like that, I would agree. The performance (and memory) requirements to get high quality rendering from voxels are extremely high. And I would add that the requirements to get polygonal rendering to the same standard are much lower in most circumstances which would go some way to explaining why the hardware focus is on that kind of technology.

 

[ector]

Delta Force, Outcast and similar aren't really "voxel" engines, they're just special-cased heightfield renderers. Wherever Outcast had overhangs or animated characters, it used polygons.

 

[Saem]

The thing with lego is that you have different sized bricks and generally speaking, only 4 orientations.

 

[MrWibble]

The thing with lego is that you have different sized bricks and generally speaking, only 4 orientations.

Which makes it considerably more versatile than voxels...

Which in turn is only one of the many reasons I'd rather play with lego than voxels

 

[Saem]

There are many relevant lessons in lego.

For instance, my nephews have acumalated a fair bit of lego for themselves, it's largely simple bricks. Most of the things they build/model are inorganic structures and they have a hard time thinking about organic shape approximation. Interestingly, enough, get them playing with k'nex, especially the ones with the flexible columns and they're making dinosaurs and what not. Very weird, consider that here, we feel things to be the other way around when it comes to their analogs.

 

[purpledog]

the only way voxels will be more suitable for phot-realistic rendering is if performance and resolution are infinite. pretty soon we'll have hardware that can render so may polygons it won't matter, because each pixel will be occupied by 1 or more polygons..

IMO, voxels are good for volumetric stuff. An volumetric stuff doesn't care too much if 1 pixel is occupied by 1 or more polygon: what's important here is that the object is fuzzy and has a very complex topoly (not like a surface).
This lead to HUGE fill-rate, because everything is hiding something else, and this is VERY hard to predict.

In those case, voxels are more likely too solve this fill-rate issue (and aliasing) by using a simpler structure (a grid), which is easy to simplified and sub-samble.