How Bump Mapping Works

The Chef writes: "Tweak3D has a pretty good article explaining how bump mapping works with 3D accelerated video cards. They cover all the basics of bump mapping and the advantages and disadvantages of several methods. Now if someone asks me how per-pixel shading or environment mapped bump mapping works, I'll have an answer (but I'm not sure if that's a good thing)." With the introduction of the new graphics cards, this is some interesting reading.

Comments

Re:An open question

[nyet]

Real "hard" core gamers turn off every special effect and level of detail to maximize frame rate, no matter how powerful the machine. All things being equal, the person with the higher frame rate (and ping) wins. I'm guessing that if the renderer actually allowed simple wire frames with basic z buffering, "hard" core gamers would set it that way.

An open question

[JosephMast]

for all the hard core gamers out there... does this sort of technology really help make better games?
On one hand I really am impressed with the impressive graphics that are apparent in the majority of today's games, and perhaps this will allow game designers to think less about the technology (even "bad" code can hit 60fps with a smoking graphics card) and work on the game...
On the other hand, I think that technology shouldn't be the focus of a good game (aka who cares if the orange is triple pass bump mapped at 140 fps if the game isn't fun to play) while games like diablo (640x480 resolution) still get dusted off and played.... any thoughts?

Re:Holy Grail of 3D Graphics?

[mav[LAG]]

It took several man-years to build, so it's not exactly something you write out of a hobby.

Don't lie Jouni - I've seen your picture in the credits for INSIDE :) And that used SurRender the mighty 3d engine too yes.

While this kind of dynamic methods are certainly more expensive to perform at run-time than pre-calculated visible sets, they are not mutually exclusive with those. You can still use portals and cell based visibility as well as static PVS if desired, but what's important that we're finally reaching a point where you don't need to

Cool - I look forward to seeing some new games with the new component.

Hair, Fractals

[Perdo]

Is hair just sharp bumps? Is there a limit how sharply convex a "bump" can be? Are we moving toward fractal geometry? big brushes for big objects, smaller brushes to add definition to the large brushes, bumps on the smallest brushes. Could fractal algorithms be used to generate complex shapes/textures without the standard 3D modeling technique of brushes + new bump mapping?

Re:Hair, Fractals

[SurfsUp]

Is hair just sharp bumps?

Nope. Once way to make hair, start with a wooden log light by a spot in the middle (this will be the sheen), stretch it out and alpha it a few hundred times onto a surface with slight longitudinal displacements and slight bending - now it should like a sheaf of wheat. Take the resulting texture and map it two or three levels deep onto a hair-shaped object, using an alpha map to give it shape and taper. On the head place an opaque map, darkened, of the same texture so you don't get any bald spots. Presto! It looks like hair.

This runs pretty efficiently in real time too, because you're just mapping a few 10's of textures to get the effect. The main disadvantage of this approach is that the highlights (sheen) don't move. A hack to make them move is to light the hair texture with a spotlight before alpha mapping it onto the hair polys.

You could add a bump map to the hair texture and you would get the moving sheen.
--

Bump mapping the nVIDIA way

[Ella the Cat]

I was in a hurry and my machine fell over as I was about to post this interesting link from the nVIDIA site. Lots to learn about therein. Sorry if I posted twice, give me a break.

The paradox of hardware acceleration..

[mav[LAG]]

..is that it becomes more and more important to find and use good visible surface determination algorithms correctly in software. When you hear figures of 3 million polygons per second done by the latest GeeWhiz 2 GFX card, remember that has to be divided by 30 frames per second to get acceptable animation quality. SGI's InfiniteReality Engine pumps out 100 000 polys at 30 frames per second. Quite quick - until you realise that a complex model of an aircraft or a city may be comprised of tens of millions of polygons. Rejecting as many of those as possible as quickly as possible (normally because you can't see them from your viewpoint) is problem which many bright people have been hammering on for 30 years now.

I know many graphics coders who are depressed because all of their hard-won knowledge coding polygon fillers, environment map effects and realistic shading engines in software seem completely superceded by advances in hardware. They shouldn't be. There's still tons left to research and better algorithms to be found - even more so now that more powerful graphics cards are becoming cheaper.

There's zillions of good Web references on the subject - here is a place to start.

Re:The paradox of hardware acceleration..

[mav[LAG]]

Probably - I got them from a paper on the subject which is ~2 years old. 208m polys per second sounds a little better - that's ~7m polys at 30fps. Does that figure include shading, T&L?

Books are better

[DrSkwid]

try readin :

"Computer Graphics : Principles and Practice"
Foley, van Dam, feiner, Hughes

Mine is second edition 1993

everything from lines to fractal hairs

inc. anti aliasing & filtering etc.

A must read for anyone more than slightly interested
.oO0Oo.

Origins of EMBM...

[Jouni]

... a.k.a. Environment Mapped Bump Mapping are not actually found in Matrox G400 but in Pyramid 3D from Tri-Tech and Bitboys, the latter more known for their long-awaited and much discussed Glaze 3D architecture. Matrox did the first hardware adaptation that ended up in the mainstream.

Some years back, in the days of Pyramid 3D (yes, boards existed!) the pixel pipeline of the graphics chip was already programmable in microcode and EMBM was working perfectly in hardware. Slowly, maybe, by today's standards, but visually as attractive as ever. While it's a shame the boards never made it into the public, they still managed to make a significant contribution to PC graphics technology.

Bitboys licensed the EMBM solution to Microsoft to make it a part of the Direct3D standard. Once it was a part of the standard, other vendors such as Matrox were also free to make their implementations of the method.

It's a hack, but it's a good looking hack. Long live good looking hacks! :)

Jouni
--
Jouni Mannonen : 3D Evangelist @ SurRender3D.com

Bitboys - a product? When?

[tjwhaynes]

Some years back, in the days of Pyramid 3D (yes, boards existed!) the pixel pipeline of the graphics chip was already programmable in microcode and EMBM was working perfectly in hardware. Slowly, maybe, by today's standards, but visually as attractive as ever. While it's a shame the boards never made it into the public, they still managed to make a significant contribution to PC graphics technology.

So they actually made some prototypes? I remember looking at the early screenshots of the Tritech Pyramid 3D cards and wondering what the frame rate was with bump mapping, reflections and fogging on.

On the flip side, I do wonder whether they will ever get a production board out of the door. The are notorious for not producing.

Cheers,

Toby Haynes

BM can be fun in software, too

[Stephen Samuel]

I remember seeing Bump Mapping done in software back in 1992. I still have a video tape of the animation (each student group was allowed 450 frames on a write once video laserdisk. It was also quite interesting seeing what people could do to stretch out 450 unique frames).
I'm actually impressed that it took this long for the concept to go from software to hardware.
--

When OpenGL?

[Anonymous Coward]

Since the introduction of Matrox G400 (see the fantastic shots at http://www.matrox.com/mga/feat_story/mar99/slave0_ scrshot.htm) Environment Mapped Bump Mapping has been a must.

The big issue is when OpenGL, and thus Q3A et al., will have it. Today it seems to be an M$ ballpark only; with the help of Bitboys' patent.

Heres the proper link

[ghoul]

here