John Carmacks "100 passes per polygon" achievable

[Brimstone]

John Carmack mentioned

In a recent statement, id software's John Carmack said that he didn't want more polygons. Instead, he wants "100 passes per polygon." (With each rendering "pass" a computer makes before it displays the graphics, more and more detail can be added.)

Carmack is no longer demanding more polygons, he wants smarter polygons. He wants the ability to bump map—simulate raised textures on the surface on polygons—and to add sophisticated lighting effects. By adding more rendering passes, designers will be able to polish their creations on a pixel-by-pixel basis.

"You had two passes per polygon for the longest time," explains Kevin Stephens, director of engineering at Monolith, the company whose advanced LithTech* engine powers No One Lives Forever 2* and the upcoming Tron 2.0.* "You got one pass for the base texture and then another for a light map."

"Newer technologies can do more. It allows you to do light mapping, bump mapping, and all of the different rendering techniques that, in the end, make objects look more real."

Stephens goes on to predict the direction engines will move in the future: "One hundred passes might be a little bit extreme, but 10 or 20 passes per polygon will not be uncommon. Having the additional passes will make games look like they have more polys even if there are no additional polys." The result? Organic, realistic shapes and objects that have textures so real you feel like you can reach through your television or PC monitor to touch them.

http://www.intel.com/personal/do_more/gaming/stories/game_engines.htm

So any guess to how many "passes per polygon" Microsoft next-gen hardware will be able to do.

 

[Fox5]

Bleh, John Carmack is always demanding stuff that he never gets, and then he forgets about the old stuff! What happened to 52(or was it 64) bit color that he wanted?

BTW, I thought bump mapping didn't rely on texture passes and was applied before the textures were?

 

[Tagrineth]

Bleh, John Carmack is always demanding stuff that he never gets, and then he forgets about the old stuff! What happened to 52(or was it 64) bit color that he wanted?

What do you think FP16 and FP32 are?

Hint: the 16 and 32 bit parts are per component.

BTW, I thought bump mapping didn't rely on texture passes and was applied before the textures were?

What do you think a bump map is?

 

[jvd]

can't you already apply up to 6 passes per polygon on the neon 250 or was it the kyro .

 

[Fox5]

Bleh, John Carmack is always demanding stuff that he never gets, and then he forgets about the old stuff! What happened to 52(or was it 64) bit color that he wanted?

What do you think FP16 and FP32 are?

Hint: the 16 and 32 bit parts are per component.

BTW, I thought bump mapping didn't rely on texture passes and was applied before the textures were?

What do you think a bump map is?

Ah I don't know, I just read something an article at Toms Hardware not too long ago from when the geforce 3 came out, and to me it sounded like the memory expensive multi texturing would no longer need to be used and instead the changes were applied during the creation of the polygon and required almost no memory.

can't you already apply up to 6 passes per polygon on the neon 250 or was it the kyro .

Wasn't aware multi texturing was a strong point on those cards(I thought they may have been a 1x1 or a 2x1, but possibly a 4x1) but I think you can do up to 8 passes on the current radeons and geforce fx's.

 

[Vince]

Re: John Carmacks "100 passes per polygon" achieva

Don't get me wrong, Carmack is the proverbial man (all hail John) but is this really what we want? It would seem to imply a somewhat static advance in the actual gameworld which are built in the same ways, just look prettier due to the perception.

Personally, I think game worlds need to become vastly more immersive and large in scope, not the same corridor or 'virtual' corridor shooters with 3-5 characters on the screen at most. Or maybe I'm just a sucker for things of epic size and scope.

 

[jvd]

Sorry its

8-layer multi-texturing support

from the kyro cards

 

[Paul]

but is this really what we want?

For enclosed fps games such as DOOM3? Yes, remember, you can use different rendering techniques for a different type of game.

Personally, I think game worlds need to become vastly more immersive and large in scope, not the same corridor or 'virtual' corridor shooters with 3-5 characters on the screen at most.

Agreed, though there will always be games like doom 3 that are enclosed. What I would like to see is a greater emphasis on polygon power and not pseudo stuff like bump mapping.

 

[jvd]

Agreed, though there will always be games like doom 3 that are enclosed. What I would like to see is a greater emphasis on polygon power and not pseudo stuff like bump mapping.

Why . I'm a fan of both . Give me faster bump maping and more polygons .

More is allways better .

 

[akira888]

Neon 250 (DC GPU) is a 1x1.
All Kyros are 2x1.

Radeons 9500+ and Geforce FX can sample 16 textures per one pass. Radeon 8500-9200 can sample six, Kyros eight, the Geforce 3-4 four, Geforce 1-2 two, and the first Radeon could sample three in one pass. In the console world NV2A - 4, Flipper - 8, PS2 GS - 1. A "pass" simply refers to what can be done with one sending of geometry, not in one clock cycle. A Radeon 9700 Pro can sample 16 textures in one "pass" but sixteen clock cycles (best case).

I have a feeling that what Carmack means by "pass" is not a "pass" in the canonical sense but rather an arithmetic blend op or a texture sample - basically he wants really long shaders.

 

[Steve Dave Part Deux]

Flipper can actually sample 16 textures in one pass. However, it is restricted to 8 unique samples which can be accessed twice each, I believe.

 

[Qroach]

There's only so much more you can do when the polygons get down to sub pixel levels (specifically with occlusion culling). At that point you'd be better off spending processing time on render passes.

Regarding consoles, I'm sure Guden will argue that the extra fillrate and such isn't needed but in reality, if it's there (at least in a console) it will end up getting used.

btw, anyone that thinks Carmack asks for things and deosn't get them is totally wrong.

 

[akira888]

Flipper can actually sample 16 textures in one pass. However, it is restricted to 8 unique samples which can be accessed twice each, I believe.

Similiar to Radeon 8500-9200, six unique textures but two different accesses each (one in each phase).

 

[akira888]

Re: John Carmacks "100 passes per polygon" achieva

Personally, I think game worlds need to become vastly more immersive and large in scope, not the same corridor or 'virtual' corridor shooters with 3-5 characters on the screen at most. Or maybe I'm just a sucker for things of epic size and scope.

Well, that's a seperate issue. I understand and agree with your desire with regard to game design, but improvements in programmable fragment/vertex/surface shaders will help just as much in Xenosaga 4 or EverQuest 3 as they will in Unreal Tournament 2007 or Quake 5.

 

[bbot]

Compare what Deano Calver has to say with what John Carmack said:


" One thing working with (and liking) the PS2 architecture is that you really appreciate polygons, on the PC side of things lots of people talk about low poly with per-pixel lighting being the way forward. I have to disagree, I want high poly with per-pixel lighting icon_smile.gif
Low poly work looks horrible, having good lighting and shadows doesn't componsate. A character (that undergoes skinning and animation) needs at least 5000 polygons (10000 is better), else the animation looks terrible and animation is more important than lighting. The whole shadow volume/per pixel lighting movement is producing some horrible looking characters compared to the low-tech PS2 characters. On PS2 they are approaching high number of polygons for characters (IIRC Jak and Dexter 2 has 12000 polygons) which mean muscles vains etc. Normal maps are better than no geometric details but they aren't a replacement. "


If John and Deano were to get together, I can imagine the following exchange:

John Carmack: "More rendering passes!"
Deano Calver: "More polys!"
<repeated many times>

 

[jvd]

Well deano seems to be playing towards the strengths of his platform. And is basicly saying how great it is . He can't push the ps2 and claim the same things carmack is

 

[akira888]

Most of Dean Calver's recent projects have been PC based.

 

[PC-Engine]

If JC's comment assumes we have reached the point of subpixel polys then yes he's correct and Dean's comment doesn't apply.

 

[MrWibble]

I disagree slightly with what Deano says. I think a good lighting model and top-notch shading are easily more important than more polys. Animation is also important, but I'm not convinced that you *need* 5000 polys in a character to make it look real. (Besides which we're already in the realms of having that many polys in a character)

Environments especially can be allowed to look a bit boxy - a lot of the real world is essentially made of boxes anyway (and a lot of it isn't, but boxes go a long way). If you light them well it creates a far more convincing scene to the average human eye, than if you slap a photograph of a crate on the side of everything.

No, for once I think I have to side with Carmack, I'll take more processing power over more polygons for the time being. Once we crack a few more issues with lighting then I'll worry about the details we have trouble representing without polys.

The 100-passes might be a bit misleading though, as with pixel-shaders we're getting into the realms of doing that much more work over a standard textured blend operation anyway.

 

[London Geezer]

Ok, so basically we should stay at Doom3 level of polygons numbers, with even more fake looking bump maps?

Wow, we're curse to play games where monsters have octagonal heads, but at least have good lighting!!

WHEN we reach the point where we have so many polygons they won't fit on our screen, THEN (and only THEN) we might start to think about 100 passes...

Personally i think Displacement Mapping is the way to go, but who the fuck am i to talk...