John Carmack PCGamer interview...next engine to have "unique texturing on everything"

[zed]

zed JC has specifically stated the procedural texturing has always been mentioned as the next big thng in "consoles" and other generational hardware but it never works.

whilst this is true in the past, in the future with multicore, u have the power that will otherwise be sitting around idle. esp the ps3 + future many multicore pcs ok since the main game code is linear u have to do it on one core, what r u gonna use the other 6 cores for? AI/physics! sure but not all 6 cores worth
anyways heres a post along similar lines i done a while ago
http://www.gamedev.net/community/forums/topic.asp?topic_id=364647

 

[KimB]

The problem with procedural isn't so much in terms of processing power, but rather on the content creation side of things.

 

[Mate Kovacs]

Now, here's what I'm going to propose is going on. Let's imagine that you store the full 32k x 32k megatexture in 256 x 256 blocks. It would be relatively easy to pull out the nearest four blocks and pack them into a single 512 x 512 texture. Then, just do the same exact thing for the next three MIP map levels (16k x 16k, 8k x 8k, and 4k x 4k), always ensuring that the nearest four 256 x 256 blocks are stored in video memory as a 512 x 512 texture each.

Doesn't that mean you only gather information from a (roughly) 4096^2 texels around the viewer? (Since a texel at the (4k)^2 MIP level represents 8^2 texels from the (32k)^2 MIP level, so a 512^2 block at the (4k)^2 MIP level represents (512*8)^2 texels from the (32k)^2 MIP level.)

 

[KimB]

Doesn't that mean you only gather information from a (roughly) 4096^2 texels around the viewer? (Since a texel at the (4k)^2 MIP level represents 8^2 texels from the (32k)^2 MIP level, so a 512^2 block at the (4k)^2 MIP level represents (512*8)^2 texels from the (32k)^2 MIP level.)

Right, so clearly this is a limitation. Either you'd have to store more 512x512 blocks in video memory (down to the 512x512 MIP level of the original 32k x 32k texture), or just stand by the assumption that anything beyond the local 4k x 4k will not be visible anyway.

An alternative solution would be to only store every other reduction (i.e. store 512^2 maps from the 32k^2, 8k^2, 2k^2, and 512^2 levels).

 

[RejZoR]

Carmack is really The God of 3D graphics.

 

[Arun]

Carmack is really The God of 3D graphics.

So, does that make Deano/nAo the anti-gods - the ones who own Gods?

Uttar
P.S.: I'm not saying they're the best (graphics) programmers in the industry, I don't know enough of them to make such a bold statement, but saying Carmack is the absolute best nowadays is ridiculous. He's still great, but far from a god, at least relatively to the "competition" he got in that domain.

 

[Richard]

Ideally, a full Megatexture toolset would come with very robust features for painting these effects on, such as procedural generation of all sorts of textures both as the base set, and for painting on (imagine just painting a roadway onto the terrain, for instance). Of importance for the road, for example, would be some elevation of the geometry that is under the surface of the road, as well as a realistic blending between the surface of the road and the surrounding terrain. For very hilly terrain, you could even flatten the terrain as the road is painted.

*nods* great tools would be a big help, and ET: QW is supposed to come with some new terrain/MT tools but ET:QW is being created by Splash Damage which have their roots in the modding community. id's tools have always been powerful but obtuse so I'm not too confident.

Perhaps JC's comments about the xbox 360 being a better platform because of better dev tools and documentation compared to the PS3 will force an improvement on id's own dev tools?

 

[zed]

The problem with procedural isn't so much in terms of processing power, but rather on the content creation side of things.

true whilst somethings are easier done than others, everything benifits theres been papers on procedurally generating trees/buildings etc. take prolly the apex of CGI a humanbeing
im looking down at my arm, with its thousands of similar hairs (ild guess in the region of a million on the whole body ), its wrinkles or splotches. even with terrabyte cards/computers u still wont have enuf storage capability to model a scene with this detail. the only way is with huge compression techniques/ procedurally creation etc. procedural stuff IS the future there is no escaping it, the question of course is when, i say now with multicores its time has finally arrived.

but saying Carmack is the absolute best nowadays is ridiculous. He's still great, but far from a god, at least relatively to the "competition" he got in that domain.

based on the evidence (and what else can u judge by ) carmack still remains in the top echelon of programmers, why ppl knock his ability escapes me, true he gets more press than perhaps warrented, but is it his fault? no, its the medias. he certainly doesnt appear to be someone who goes out seeking it. ( ill change my opinon though if he shows up on a reality gameshow )

 

[Laa-Yosh]

Anyone talking about procedurals should look at Pixar's stuff, particularly Bug's Life, because that is how it can look like. Creating anything remotely realistic with procedurals is going to be very, very hard, if not impossible; things like metal surfaces, human skin, ground in close-ups, and dirtying and weathering and such effects are usually problematic to get right. They'll always look like a fractal texture, too CG-ish even for the average viewer.
Procedurals are good at fire and smoke, and at landscapes viewed from higher altitudes (see Terragen). For almost every other case, painted textures are second to none and that is why photoreal/hyper-real movie VFX stuff is still based on that and not procedurals.

 

[KimB]

true whilst somethings are easier done than others, everything benifits theres been papers on procedurally generating trees/buildings etc. take prolly the apex of CGI a humanbeing
im looking down at my arm, with its thousands of similar hairs (ild guess in the region of a million on the whole body ), its wrinkles or splotches. even with terrabyte cards/computers u still wont have enuf storage capability to model a scene with this detail. the only way is with huge compression techniques/ procedurally creation etc. procedural stuff IS the future there is no escaping it, the question of course is when, i say now with multicores its time has finally arrived.

That's all well and good, if anybody figured out a good way to procedurally generate that content. To date, procedural textures have only been useful for simulating a very limited set of surfaces.

 

[Farid]

Anyone talking about procedurals should look at Pixar's stuff

The thing is Procedural Generated Content doesn't necesarilly imply that it's 100% "algorithmic" based.
You can create procedural content that uses bitmaps, and hand created model, as a part of the generated content.

Just like what some games are actually doing. It's not as memory efficient as what full procedural generation is, but it helps a lot nevertheless. Some games on PS2 actually do that. It pushes the polycount sky high, it doesn't have the habitual "procedural" look to it, and it's, of course, a lot easier than classical PGC to work with for the art team.

And LOL at the "John Carmack is the guru of all what's 3D".

 

[Junkstyle]

So can Carmack just come out and clarify what new stuff he's bringing to the table? It sounds like this ***MEGA-TEXTURE*** technique is pretty common.

 

[Sxotty]

I think you are all taking the procedural comment out of context. Carmack it seems to me thinks that dynamically doing procedural content without intervention of a human is silly.

He is sayig that they do use procedural routines to generate content, and then paint it and tweak it to make it look nice. This achieves the goal outlined previously: Namely to reduce the workload on artists. And I do like some parts of Cramacks style in general, but he is pretty inflexible on his attitutdes I admit that.

 

[Blazkowicz]

So can Carmack just come out and clarify what new stuff he's bringing to the table? It sounds like this ***MEGA-TEXTURE*** technique is pretty common.

He didn't invent stencil shadow or DOT3 bumpmapping either. Nor do UE3 or cry engine bring any really new stuff. Sure, you can boast about your supporting of parallax mapping and whatever but every other engine will have it as well.

So I guess it's more about the implementation and how you use the features rather than features themselves. (Carmack's reverse and normal mapping everything for instance)


that said I'd like to know how well will run quake wars on my geforce 4!
with every new carmack engine people feel it's time for upgrade and bitch about it and resell their wildly expensive hardware to buy even more expensive hardware..
But as quake 3 based games run fine on a voodoo2, doom3 runs as fine on my gf4 (which is much better than say an X300SE for that). Keep in mind doom3 shadowing and lighting was made to run on NV10.

Carmack does wonders to get the best of what so quickly becomes "outdated" cards and I respect him much for that.

 

[zed]

just had a quick seacrh came up with this for procedural skin
http://www.shonner.com/drafts/carrara_skin_tutorial_a.htm
not to bad, whilst if u refrain from letting the camera getting to close to the subject, procedural will never be as good as idivually tailored texture maps. having unique textures on everything is impossible. eg take the latest GTA game where u have 100s of characters running around with skin like http://www.shonner.com/drafts/images/athena_back_skin_test_12.jpg to have them indivually textured to such quality wont even fit on a double sided blueray disk (never mind the cars, building, terrain etc)

Sure, you can boast about your supporting of parallax mapping

noone boasts about parallax mapping do they? thats embarrassing, any engine that boasts about that as a standout feature is limitted

 

[KimB]

noone boasts about parallax mapping do they? thats embarrassing, any engine that boasts about that as a standout feature is limitted

Why? It may be simple, but it's quite a powerful technique for improving the quality of bumpmapped surfaces.

 

[Richard]

noone boasts about parallax mapping do they? thats embarrassing, any engine that boasts about that as a standout feature is limitted

The UE3 page boasts of parallax mapping and "limited" is hardly the qualifier I'd use to describe the engine.

 

[zed]

The UE3 page boasts of parallax mapping and "limited" is hardly the qualifier I'd use to describe the engine

i dont see it here
http://www.unrealtechnology.com/html/technology/ue30.shtml (is there another site they use to discribe the engine, ild like to see a link thanks)
but anyways i think u misunderstand me, i mean 'boast' as in showing off what it can do as special, ie standing out from the crowd,
parrallax is extremely simple to implement (a couple of extra funcs in the shader (perhaps also extra texture access), plus also height info in a texture made by the artist, it comes at a very low performance cost ). im pretty sure u can get any existing engine (which supports bumpmaps etc) supporting parallax mapping with 1/2 hour.
i assume unreal has a huge list of things it can do (parallax being listed along with say bumpmapping) i wouldnt say it boasts of it.
im not dissing parallax at all, the cost vs quality ratio is very good, i just wouldnt lump it up there with a unified lighting pipeline etc.

 

[Richard]

i dont see it here
http://www.unrealtechnology.com/html/technology/ue30.shtml (is there another site they use to discribe the engine, ild like to see a link thanks)

That's the one; look where they mention "virtual displacement mapping" which is what Epic calls parallax/offset mapping.

 

[AlNom]

It was also the first thing they showed in their materials tech demo from way back. They used a brick wall to demonstrate.