Sparse Voxel Octree (idTech6)

[obsidian]

idTech 5 isn't even out yet, but even then, id Software has been researching potential technologies for idTech 6, one of which John Carmack mentioned in one of his QuakeCon keynotes (I think this year) about sparse voxel octree. SVO basically brings unparallelled geometric detail into the game engine by replacing polygons with "voxels" (volumetric pixels), throwing out the need for normal maps, parallax displacement maps and all that other fluff.

You can think of voxels as being little individual 3D pixels that makes up the model. Cool thing is that you can stream this stuff to the renderer, much like how megatextures are being streamed to contain pixel data, only voxels are streamed to create 3D data. So you can zoom in on details infinitely (or as high as your reference model is) and you'll still see every single detail without loss to performance when you zoom out and view the whole model. No need for LoD, either.

Seriously, this is next-next-next-gen stuff. Here's a demo video by Jon Olick from id Software at Siggraph 2008 demonstrating with a 3D model with a source model of 7 million polygons "converted" to voxels rendered from a single nVidia GTX 280 GPU at 60FPS. It's been done in 3D games to some light extent, but certainly not on this level... he zooms in on details on the character model's FINGER!

[Kat]

From what I understand about this it's only usable on 'static' (?) objects like environment assets and such like?

I'm in two minds over this stuff now because modding is going towards needing more and more applications in order to 'compete' at the higher end of things, making normal maps from 2D artwork is poo'd on as it is so once voxels come into play you really are going to need to use sculpting tools to be productive (and/or think about employment). Is stuff like this going to handicap the creative community/pool of talent?

Upside is we could potentially finally start getting better terrain!

Sorry for taking this slightly OT on the negative(ish)

[^misantropia^]

Everything old is new again. I wrote a small 3D game in 1995 that rendered voxels (in software) to generate landscapes. I think most flight sims do something similar. Though admittedly this is a rather clever and sophisticated application of the technology, what with it retaining lighting info, normals, etc.

[o'dium]

You can see dirt in a player models finger nails... It doesn't mean it makes a great game.

id Software of late have lost the old greatness because they seem to be chasing the next big thing, while skipping out on noticing that the "next big thing" gamers want is... Game play.

Dont get me wrong, that tech video is amazing, really is. But as Kat said... Who is gonna do it...? The current work load for OverDose is tiny compared to this, and we struggle with it. Do they really think that many people will do it...?

I'm all for these tech videos and stuff. Its great seeing what my childhood heroes are up to. But hell, I would KILL to see them release a tech video showing off new AI in a game thats so realistic and adaptive it makes you think twice if they are human or not. Who thinks that no matter how good this model looks, it will still run right towards you and take 50 shotgun blasts to the head to kill...?

I just want a dev company now to say "you know what? Games look good enough. Lets move onto what makes the gameplay better..."

[Hipshot]

Delta Force, Outcast - great old games using voxel tech =) Fun to see it's on the topic again. Some people I've talked to in the biz, here at work, speaks of voxels as a bad thing, I guess they see it as old and not usable any more.

For myself, I don't really care about how advanced the graphics are on a tech base, more how it looks from Art Direction and style, I guess that's why I still like so many old games and prefer many of them over new games.

@ Odium, I would have taken you more seriously if it wouldn't have been so that everything you showed from your project is graphics-only related things... =)

[o'dium]

I cant show gameplay in screenshots chap

But yeah, I knew people would say something like that lol, I was asking for it

[obsidian]

Stuff

The grown ups here are talking about the technology (SVO), not video game fan nerdism. I appreciate your comment about the potential increase in workload for freelance mod work like on your project, but where the hell does all this fluff about gameplay and "blah blah blah cry about id Software" stuff come in? Feel free to post again when you've grown up.

[o'dium]

Grown up? What, just because I didnt piss myself with excitement that id software are planning another game based on graphics and not gameplay...?

Come on chap, the point is that no matter how "nice" this looks, it will still be used in a bog standard, dumb AI first person shooter.

The post above, if you read it without acting like a twat, simply states that me, as a person, is fed up with all the new technology leaps in GRAPHICS when nobody ever seems to do next gen AI, or anything that would improve the experience.

If your gonna post a topic on a discussion forum, then you must welcome all forums of thought on that, and as long as its not off topic, you should embrace the discussion, not go off on one because I didn't post a virtual blow job to id Software.

[obsidian]

From what I understand about this it's only usable on 'static' (?) objects like environment assets and such like?

That's what it sounds like. Still a world away before they have anything really functional as a game, so I guess we'll see if they can fit everything into the same pipeline. He certainly talked about it as a strong possibility but probably leaning towards the next iteration of the technology. BTW, would you happen to know if megatextures for idTech5 support dynamic objects as well?

I'm in two minds over this stuff now because modding is going towards needing more and more applications in order to 'compete' at the higher end of things, making normal maps from 2D artwork is poo'd on as it is so once voxels come into play you really are going to need to use sculpting tools to be productive (and/or think about employment). Is stuff like this going to handicap the creative community/pool of talent?

Arguably, mod work is getting more difficult regardless and it's hard to create something on the level that you see by professionals. What it does do is free up resource budgeting on the professional level, so artists don't need to worry about how much detail they can put in before performance starts taking a serious hit. You don't need multiple instances of models for LoD, you don't need to render out normal maps (yay - if you remember my ring model for the knocker, normalmap aliasing sucks), you just export the model and you're done. But yes, you may need to learn to use sculpting tools for stuff like terrain and organic objects.

Oh... another thing... since voxels contain colour info as well, I suppose that means you may not need to UV-map your models, you can paint them using 3D painting tools and maybe export directly to some format for rendering. Fingers crossed for that... UV-mapping is the most hated part of my workflow.

[obsidian]

More stuff

Just pretend... just for a second... that we're not talking about something made by id Software. Just pretend that we're only talking about SVO's for game development and come back and say something intelligent.

[o'dium]

Even then... Why is it more important to see the wrinkles on a models finger? Why does a team not make some awesome AI, or some better more realistic way to interact with the game world...?

Just pretend for a second I have a point

[obsidian]

Talking to o'dium is like talking to an infinite loop.

[GONNAFISTYA]

No...it's like talking to a brick.

[Grenader]

No...it's like talking to a prick.


Anyway, looks amazing, can't wait to see how it will look in real time though. I can just imagine actually being intimidated by that big scary monsters face on my screen.

[sock]

Upside is we could potentially finally start getting better terrain!

Cryengine from Crytek has a terrain grid which can be shaped with height maps or pulled around with special terrain brushes, has a terrain painting system which does all the blending for you (16 base material types) and voxels for creating organic shapes. (Mostly used for indoor cave stuff)

Just like ID tech the terrain can be enchanced with static meshes and decal stamps. The voxels are created inside of portal cubes and can be made to blend with existing terrain edges. (Just cut a hole in the terrain and line up the voxel cubes to the hole) In an ideal world both systems (terrain and voxels) would be the same but the voxels can be made to line up with the terrain so overlapping can be done without too much trouble.

Linky to cryengine features: http://en.wikipedia.org/wiki/CryENGINE2

Sims

[Detoeni]

(Mostly used for indoor cave stuff)

and every where there's sculpted terrain, rocks, cliffs and the likes. They got a lot out of this feature and its also turned work flow on its head. Being able to wave me mouse over my monitor while muttering “make cliffs here” beats modeling them in max any day of the week.

[Castle]

Even then... Why is it more important to see the wrinkles on a models finger? Why does a team not make some awesome AI, or some better more realistic way to interact with the game world...?

Just pretend for a second I have a point

A voxel based system like the one described above has more potential to create an environment that allows for the kind of new and fresh feeling game play you desire.

The reality of this is a system that when combined with new age physics engines your next death match could involve toppling a building in an attempt to kill your opponent.

It could also potentially allow us more streamlined design pipelines. I am all for that as well.

Stop acting as though you know when and where the next big new game play idea is going to come from. Chances are pushing new tech like this is exactly what is needed to make it possible.

[Castle]

Reading up some more on how Voxels work I have to admit this system has surprising potential.

Speculating:
From what I can tell your artist creates a high poly model then when it is done it is converted into an SVO.

From what I can tell Voxels are basically 3 dimensional textures. So you would also be likely inclined to texture your high res model as well however you see fit. Once you generate the SVO it would most likely retain all of your color map information from the high poly model.

The SVO appears to be broken up into square segments. An octree. You see a voxel works a lot like a texture does in that each pixel does not have information where it is in relation to the world around it. Much like brushes in radiant you only need a starting point and the rest is figured out from that starting point.

Well in this case I imagine that the Voxel is broken up into segments to allow it to deform with out having to recalculate the entire characters body.

Things like being able to blow chunks of his body apart or tare him into pieces could be possible with out the need to create additional assets for his limbs. You could potentially deform the actual object itself if I am correct.

I haven't the slightest idea how you would rig a SVO character like that but I am pretty sure it could be done. With a really robust shader system you could actually create and move geometry with just the damn shaders..

I also speculate that with such a system as designers and artists we will continue to build our worlds and characters with polygons. Characters would require high poly models still while level designers might simply just place textures that actually deform into 3d surfaces when converted into SVO.

Madness I tell you. madness.. genius.

I wonder what the file size of that one dude is?
Anyone know more detail about this I am terribly interested.

[pjw]

Y'know, I may get slammed a bit too, but I honestly agree 100% with everything in o'dium's initial post. In fact, this:

I just want a dev company now to say "you know what? Games look good enough. Lets move onto what makes the gameplay better..."

is going into my quote file.

On the other hand, this thread is about discussing and spooging over new and exciting tech. Maybe if you want to talk about gameplay vs. eye-candy, you should make a thread about it? -Serious suggestion; not being a sarcastic ass.

Hell, I'd post in it.[/off-topic]

The reality of this is a system that when combined with new age physics engines your next death match could involve toppling a building in an attempt to kill your opponent.

This is the sort of thing that gets me all stoked about new tech.

Anything that results in less fussy detailing, and less wailing and moaning about texture budgets and complex resource/memory management messing-with, and allows for more Big Ideas like tipping over buildings and streamlining the generation of multiple LODs on varying scales within a game (e.g. dirt under a fingernail ->grime in a gutter<-->trees on a hillside<-->streets in a city), gets my vote.

Is stuff like this going to handicap the creative community/pool of talent?

Nope.

Talent and passion will never be handicapped. People with talent and passion learn and adapt. A great artist is a great artist, no matter the tools and techniques.

[Castle]

Yeah if you can streamline a good design pipeline and provide some intuitive tools with a fool-proof occlusion culling system you will have a much better platform for you to actually pay attention to game play rather than technical stuff.

In most modern 3d engines these days design pipelines are all over the place. With a system like SVO however depending on how it is done could actually make things much easier on designers and artists alike. Again, this would depend on how intuitive the system is designed however.

If all I need to do is plop a high poly character into a compiler and it generates the SVO with out a problem that could cut a hell of a lot of trouble.

[Silicone_Milk]

Ive been wondering why voxels havent been used more often after reading an article on octree texturing. Seeing the first blip about Sparse Voxel Octrees brought a smile to my face.

irrc this will be used only for static geometry. Octrees are an expensive thing to recalculate every frame for dynamic objects. Last I read, iD is sticking with traditional polygons for dynamic entities.

Ive been seeing the voxel pop its head up in quite a few places now. It's amazing to see that this old tech is seeing use once more.

[Fjoggs]

Posting in a thread I don't understand.

[dichtfux]

I hear they'll support vehicles, too. Here's a top-secret screenie:


Note the Sparse Voxel Octree (marked in red).

[Silicone_Milk]

holy crap! It supports colors?!

[jal_]

From what I understand about this it's only usable on 'static' (?) objects like environment assets and such like?
I'm in two minds over this stuff now because modding is going towards needing more and more applications in order to 'compete' at the higher end of things, making normal maps from 2D artwork is poo'd on as it is so once voxels come into play you really are going to need to use sculpting tools to be productive (and/or think about employment). Is stuff like this going to handicap the creative community/pool of talent?

I'm not so sure about it requiring more work. In fact, it looks like requiring less work to me. In today's tech you have to work the high poly to generate the normal/displacement maps, and several low polys for the LODs. This one looks like it's using a high poly directly to generate a "3d heightmap" and the model is generated in real time from it. At least that's the simple translation to 3d of what the old ROAM octree terrain did in 2d (what missantropia talked about). So the artist would have to do just 1 detailed model to generate the "volumemap" data and that's it.

In fact, all things ID have been doing lately seem to simplificate the artist job from their iDtech4 level. Of course they require much bigger and much detailed productions, but the way those works are produced are more direct and simpler.