Question about MegaTexture and MegaGeometry

That's not a very accurate summary though. This will also alow them to get rid of LoD/UVmapping/geometry limits/normal mapping, etc. How many man-months are saved by doing away with all of this?

They've already said dynamic objects will use the standard poly-based tech as today so there's nothing stopping them from using dynamic vegetation or whatever.

WRT physics, people need to understand that a physics demo != game. And developers should stop showing their physics demo as game demos too. STFU twisted is a disappointment because the physics that are fun and can be implemented into the gameplay are really no more advanced than what other games have done.

 

Check out these videos from 2007: part 1 & 2.

 

It's a complex question...


First, they don't yet have any idea about how good their tools are going to be. It's important because most voxel based graphics today rely on some kind of real world data, e.g. medical scans of organs and such. There hasn't been much research into content creation, so it may end up being better and faster then today's approaches, or it may end up requiring more work.

It is however likely that a lot of content like terrain elements, mahcinery, crates, architecture and such has to be built using the existing poly - UV - normal map - bitmap texture workflow, and then it'll be converted into voxels.
Even they don't expect any solutions to replace characters and thus their tools/weapons with voxels either.

On the other hand, the increased level of detail will require higher fidelity content that will cause a lot of extra work on its own.



Now for the ingame part, not being a coder I can't really go too far. But if (a strong if the next gen hardware can get tesselation and subdivision surfaces right then we'll have a good LOD scheme for a large amount of content built into the engines without external help; and while the lowest resolution (the subdiv control mesh) may not be as low as today's lowest LODs, the hardware may be powerful enough to just brute force it.
Then there are many rendering related issues as well. Textures are also used for a lot of stuff other than colors. Dynamic lighting is already expected in todays games. Deferred shading and lighting relies on G-buffers that won't really work with voxels. And I'm sure any dev here could go on and on.

Do I have to repeat myself here?

And the visual difference would be even more obvious as it was back then when the low resolution made everything equally pixelated.
Meaning: normal mapped polygon models and voxel based enviroments will probably look very different from each other, which may end up pretty bad.

Also, the vegetation in outdoor enviroments makes up a LOT of the scene complexity - lots of geometry and textures, high level of overdraw, dynamic shadows and self shadowing, and so on. I wouldn't expect the first Tech 6 games to go in this direction, they'll probably choose the static enviroments which the engine prefers and can showcase very well.

While I agree, I also have to re-emphasize: Tech 6 isn't a physics friendly game, it'll impose serious limits on gameplay possibilities. Not even stuff like the pre-baked building collapse in KZ2 seems likely, and who knows what we can expect on the next gen consoles in 2010+? Id will have to compete with those games, and I'm not sure if graphics alone are going to be enough.


All in all, this new approach seems to be kinda backwards to me. It's trying to make the job of the coders easier, simplifying a lot of stuff only they care about. It's only goal is to increase geometric complexity, and it sacrifices almost every other aspect of the game engine to reach this goal. The engine will once again define and thus constrain the gameplay; and to a level, the looks as well. As good as that may be, I'm not sure it's a step in the right direction.

 

How likely is that? id completely redid their tools to support MT, what makes you think they wouldn't do the same for MG?

Jon does say that if next-gen hardware exceeds their performance expectations they have a plan B which does everything, dynamic objects included, as voxels.

This is the same argument thrown against MT and it didn't pan out because just because you can have high detail everywhere doesn't mean you have to have it everywhere.

I don't know. Characters are already pretty high poly so they're a prime example of something that does not require any extreme amount of detail thrown in. Without any screens neither you or I can say whether they'll blend in with the voxel world or not. It's also interesting that HL2 is considered one of the best looking games when it released even though it had shadow maps projecting over lightmaps, which made it look wrong and because it was mostly baked lighting all the Havok integration on the Source engine couldn't stop the designers from making totally static hanging lights. You could shoot crates and pipes and the physics would take over, but shooting a lamp hung from the ceiling wouldn't made it swing at all; and people still loved the graphics.

Vegetation (or anything really) doesn't have to be dynamic right up to the horizon. Even Crysis "cheats" in this regard. However, if it turns out that way I don't think there's a new problem to solve. Even with today's basically the same polygon-pushers you have engines which handle certain games/scenes better than others.

That seems pretty definitive when we haven't yet seen a single screenshot and, at best, the first game is 3 years away, doesn't it? If the voxel data-structure allows for realistic fluids that hugs the geometry but not a lot of falling boxes, does that make it more or less physics friendly?

 

Because it's far more practical and faster to model most kinds of things in the already existing ways. There are production proven off the shelf tools like Maya, Zbrush and Bodypaint that have accumulated decades of development effort, provide a wide range of features and are available from day 1.
Id will most likely concentrate on level editing tools to create terrain and populate it with models from a library. They'll probably also do a poly to voxel converting tool.

(Also note that id has not replaced Maya, Mudbox and Photoshop with Tech 5 either, they just put together their levels from pre-existing building blocks in the new tools.)

I still think it's unlikely but it's too early to tell.

There's only one game produced with full blown MT tech and we have absolutely no clue about how much work goes into it and how much content it is going to provide. We do know however, that id has grown to its largest size so far, and they're relying on outsourcing as well. That's a hint, considering how much content went into Doom3 back then, with a smaller team, less advanced tools and a production time similar to Rage.
But again, too early to tell.

I strongly disagree - game characters are still very limited both in geometry complexity and in primary and secondary dynamics, from the skinning through rigid body, soft body and hair related stuff. A few bits of simple clothing is all we've seen so far, there are worlds between ingame and game cinematics stuff, not to mention movie VFX. Elbows, hips and shoulders still look terrible in movement. But how could we possibly think about these problems if the silhouettes are still edgy, and a lot of the detail is still 2D? Even fighting games are pretty lacking and they only have to show 2 characters at the same time.

But the point is that the enviroment will probably also have precalculated lighting and different kind of shading, going beyond the detail differences.

I'm basing my opinion on Outcast, I have to add that I've played it again just last year so the memories are quite fresh. Tech 6 may or may not be better than that, but again, I have my doubts about the former.

HL2 is a 4 years old game, as it's been mentioned in that programming forum as well. Many things changed and Source is pretty outdated nowadays. It is not really nice when a 2012+ game is compared to a 2004 game in terms of features, either.

I'm merely pointing out the lack of flexibility in the approach, compared to what we can achieve with polygons.

Note: Please don't treat my posts as attacks or bickering, I'm merely expressing my doubts and bringing up arguments to support them.
Now, is it or is it not possible to recreate the collapsing building scenario from KZ2 that I've mentioned with voxels? Or is it possible to continously change the time of day as in GTA4? The answer to both is very likely no. Voxel based engines, particularly with this implementation, are pretty limited.

All in all, it seems as if id has found the ultimate hammer and now everything looks like a nail related problem to them. In fact the entire approach feels very typical of Carmack - I can imagine how he's been originally experimenting with octrees and high poly meshes, then suddenly realized that he could leave the polygons out altogether and still get lots of geometry, but a lot faster...

But I believe that this approach is very limited in what can be achieved - and not just with graphics, but more importantly with gameplay as well. Who's gonna licence it, how many different kinds of games will id be able to build upon it, how is unlimited geometry going to fare against GTA5s and Wii2 Sports and KZ3s?

It's like a highly advanced Formula 1 engine, and as good as it may turn out to be it's overly specialized, unsuitable for most devs' needs and most games' requirements. Thus it seems like a step backwards. But I'm starting to repeat my previous posts, so I'll leave it at that.

 

That's only because of his expectation of the next gen consoles hardware, not the limitation of voxels. On that thread even he said that dyanamic lighting is on the table if the consoles hardware is capable of calculating the lighting.

Anyway SVO is just a way to store all those detail. Their next engine will be mixed. So you have SVO for all the things that is static, much like how MegaTexture and most texture in the world is static. If they want walls to crumble or crates to tumble and break it will probably be subdiv + displacement mapping to match the detail.

I think they are betting that this solution will be faster thus providing higher detail than going all subdiv + displacement mapping.

In term of content creation, I think it will be the same with higher fidelity. Content can be converted to voxels just like the demo. Maybe you can add more detail once that is done.

I think the biggest hurdle to MegaTexture and this is storage space. The content need to be high enough resolution for at least 720p and everyone is expecting 1080p next gen. I am not sure 32kx32kx32k is gonna cut it.

 

Just one thing about id's size... they are handling multi-platform versions of RAGE internally, they decided to go with a second internal team to handle DOOM IV (and have people working on QUAKE LIVE), etc... they have a lot of projects on their hands and that pushed for the increase in work-force rather than Tech 5's complexity .

 

Actually I would quess that deferred rendering of dynamic lights is the only way to go.

Biggest problem would be the storing of attributes needed for materials, not the actual creation of G-buffer as they already have the world location, albedo and normals for all voxels.

 

The multiplatform versions of RAGE have little to no connection to the art production staff we're talking about here.

 

it'll be possible if the game world is so constrained that 32K is sufficient to describe the environment. if i didn't quit my development career i would have definitely implemented an iteration of MegaTexture to play around and learn.

 

The decals are more an implementation detail than the actual solution ... the actual solution is 3D painting (as they say, for best performance they have to bake it right back into the megatexture).

PS. I still think allowing artists to perform temporary mappings of the megatexture to a virtual 2D texture using a orthogonal/cylindrical/hemisphere/whatever projection surface is a really good idea for a 3D painting program (or do they already allow this?). Artists need to get away from both vertices and texels alike.

 

Ether (and everyone else), take a look at this if you missed it.

The podcast is mostly from the artist PoV but there may be little nuggets of info you can find.

 

Creating a 3D brush which directly smoothes the normals/textures/etc across the seam in the megatexture is perfectly possible without the decal ... the decal is just a shortcut which makes life a little easier, allowing the merging with the megatexture to be done offline.