[PC] RAGE

Yeah, you need to define what part of texture will map to what polygons. Also 1 texture alone gives uniform texture resolution but if the uniform detail is low then it is low for all surfaces.

 

What I think is id's level editor is really good and can do all sorts of fancy tricks, and people are attributing all those tricks to only having 1 texture and not to the editor

 

No, this second part is not true. Imagine the "single texture" not as a 2D image but as a mipmap tree, starting from a single pixel at top, and getting 2x more detailed at each level.

This tree is "sparse"; it stops at, say, 16 pixels per meter for the rocks on the top of the canyon where you can never climb, but it can go all the way to 128 pixels per meter for a picture hanging in the desert bar. When an artist decides to add detail somewhere, he extends just that part of the mipmap tree with more detailed leaves; he doesn't automatically make the entire world 4x more detailed.

This allows you to add detail just where it's needed, and to truncate the tree to fit hardware limitations (e.g. the DVD size on 360, or in the iPhone game).

 

A completely automated texture atlas opens possibilities simply unavailable with traditional texturing.

AFAICS the way forward is a system where the 3D editor adopts the standard layered approach of 2D texturing, with artists only ever accessing the texture through temporary projections (this doesn't necessarily mean 3D painting ... they could place a projection surface in the 3D editor, plane/tube/box/whatever is convenient, to get a 2D texture of all the layers to photoshop after which it gets reintegrated with the true texture). Once artists are once removed from the native texture automatic UV unwrapping becomes easy, the editor doesn't really care how many seams it has to work across or whether there is texture warping (except for wasting texture resolution, megatextures help a little here too by allowing variable texture resolution).

Megatexture is the first important step, the rest follows naturally IMO.

 

My point isn't that Megatexture doesn't simplify the texturing pipeline. I reckon that it does just that. What I'm arguing is that now that artists can work freely anywhere in the map without thinking about a fixed number texture set they'd have to create, they would and spend the time they're given to add details. But that detail wouldn't necessarily be 100% original textures on every surfaces (like Megatexture could allow, but disk space issues and artwork time budget couldn't afford). It would just be variations, decal effects and tweaks.

But those variations and decal blends are already used on other engines. Not to the extend to which a texture virtualisation engine like Megatexture could, of course. But as I said, storage issues and artwork creation budget realities limit what can be done. Artists can now work differently, but not necessarily much faster, since they're given new toys to play with. They'd spend less time with UV issues and more scouting the maps to tweak everything in sight. And that's the only point I argue, here.

So, yes, it's a positive change for texture artists, but the budget time for texturing of 10 would remain 10 with megatexture and the artwork wouldn't display much more originality on the base textures than on other texture streaming technologies. They'd be more variations and the texture artists would have enjoyed working on the maps much more, I'm sure, though.

That type of technology is definitely making the life of texture artists simpler. But it doesn't alleviate any and every issues they could have. They still have a finite amount of time to produce and map the textures, they still have to consider the technical issues at hand (maximum mipmap size, amount of variation in the base maps used since they can't afford to have discrete maps on every surfaces, etc.)

There's also the fact that Megatexture, as a virtual texture system implementation, imposes limitations on the whole rendering engine (filtering issues...). But that's another topic, I guess.

 

There's something very important that you should not forget here. Today most games have a sort of a separation line in their assets, where something is either almost completely uniquely textured, like characters and weapons and most vehicles; or it has to use tiled texture sets and blended masks decals etc. The distinction is mostly based on size, even though texel density isn't constant so some larger vehicles can still use unique textures if they stay relatively distant from the camera.

So this basically means that for example a building have to be assembled from a limited set of components, as there's no room for more than a few unique pieces. Sometimes artists will use tricks like rotating the same component, or using only half of it, but the general look will remain to be very organized and artificial. Sure, it isn't too far from the real world where most architecture is also based on repeating patterns, but it's still a limitation on the artists and it can ruin the immersion in many places.

Rage does not force this separation and any piece of the environment can be completely uniquely built and textured. Obviously id's not going to do it every time with every piece of rock and canyon, but when they go for it, it really looks different from everything else.

 

especially this gives consistency.
if you launch any id game on that piece of crap you have there you know it will run fast and stable and look the same way whatever the map or place of the map you're on.

now from that single screenshot you can tell there's less of separation between the characters, vehicles etc. and the environment, it feels more seamless and less like a collection of ragdolls in a sandbox of placeholder wall textures.

congrats to giving us something to look at : various copper and rusty things, leather, tired wood, bolt and nuts etc. all part of some architecture or clothes or weapons, items etc.
that's a good change from the muddy waters and moldy stones or dark labs after dark labs you expect from id.

shadows are weak you'll have to bruteforce them on high end GPUs in doom 4 I guess.

 

It's a 60Hz shooter, with no world war or modern combat theme, with lots of color, so it's already something pretty unique It also seems to have refreshingly old school gameplay, we'll see how well that works on consoles.

 

So in the same amount of time, the artist can create a more detailed, convincing world without worrying about the constraints they would normally have to deal with. Or is that not the case?

 

sorry I still dont understand how only having 1 texture means you have a limited set of objects I can certainly create a level in unreal where every object in it is unique and every thing is covered with a unique texture
also dont gfx card have a limit of 8192x8192 for a single texture ?

and for example if an artist had a cool method for creating a metal texture in photoshop that he wanted to use on a gun barrel he couldnt do it he's stuck with whatever tools and effects id's level editor has

 

In any conventional engine you're limited by the hardware platform's available texture memory. That's not much on a current generation console, and you already have some of that memory reserved for characters, weapons etc. so there's just a very limited amount remaining for the level itself. Even if you can theoretically do anything in Unreal engine, eventually you'll run out of RAM, which will limit level size and texture variety.
You can enlarge the actual size of the level if you re-use textures. Here's where you can do two things: on large surfaces you can tile textures and use multiple layers to hide the repetition, maybe even add some debris and such to cover it up; or use many instances of smaller objects like columns, doors, windows etc. This will allow you to build more with the same amount of assets, but the repetition can't fully be hidden.

With megatexture, you can uniquely texture map every square meter of a level, so you're not forced to reuse textures and objects, the complexity limitation is considerably reduced (geometry still requires memory, but for ingame lowpoly stuff it's usually far less then textures).
Also, of course the artists are completely free to paint their textures outside the level editor, it's completely silly to suppose that they're always limited by the editor. Every character, building, prop and whatever can also be built in Maya and Photoshop and whatever; the editor is an additional tool to speed up the work on the landscapes and apply an additional level of refinement with the decal system.

 

Anything to make memory use more smart is a major plus because no matter how much memory the machines have they were always be horribly constrained. I mean here we are with consoles that have 512MB and we're still calling them horribly limited when 10 years ago it was the same deal but with 4MB RAM.

I'm wondering what kind of streaming data rate the engine needs to do for megatexture. When does that become a bottleneck? Streaming off a DVD is pretty nasty for sure but I also wonder about HDD limits. On PCs streaming has been troublesome because whenever HDD accesses occur we tend to get stuttering...

 

so the artist would have to load this huge texture into photoshop and how does he know what part of that texture is going to end up as his gun barrel
and I dont understand why if you were to use 2 textures all those fancy megatexure tricks would suddenly stop working

 

You build the asset as a standalone texture and model and then you import it and the tools take care of it, I imagine. Automatic atlas mapping and all, it's getting a bit too technical from here.

 

I don't think it does - there's no reason it should, really. The thing is, you won't HAVE TO with megatexture, as you can just pre-bake those two layers together into one, thus reducing memory requirements and increasing rendering performance.

 

when i say 2 textures i dont mean 1 texture laid over another i mean 2 separate textures like the megatexture then for example a gun barrel covered with a separate metal texture
and what about mipmaps no way in hell can a gfx card generate a mipmap for a texture that is (for example 64kx64k pixels in size)

 

The graphics card never has to deal with the megatexture, the engine hides this from the hardware completely. All the GPU gets to see are the individual tiles, which are precalculated and stored in a compressed format.

You see, the very problem is that conventional engines always load the highest MIP level, even if it's only the lowest that will be used in the rendering. The entire point of virtual texturing is to only load the lowest possible amount of texture data to render the scene - so the MIP levels are probably precalculated and included on the discs.


Also, the megatexture includes all the character, vehicle, prop and other textures as well. I don't really get what you mean with anything covered by separate textures. Virtual texturing has very little to do with the traditional concepts of standard texturing, but it's still all a software layer that works on the existing hardware.

 

Also, I would suggest to do a search in the forum threads to read up on the tech before asking further questions, because it seems you don't really know much about the general concept and that's the main cause of your confusion here.

 

well your right that i dont understand it
I guess the point im trying to make/find out is all these fancy tricks like a texture atlas, virtual textureing ect do they work on multiple textures is virtual texturing limited to a single texture ? is a texture atlas ?
if no then whats the advantage of a single massive texture ?
wouldnt it be better to have multiple textures then when all your texture work is complete combine them all into a massive single texture.

 

Did he not just say that is what they do? Many textures and combine them into one.