do true circles and arcs? I was just lookin' at pictures of Sonic on xbox 360 and all his "round" features are still high-polygon-count, angular shapes. and GT4, for example, all the corners and stuff are jagged, etc. I cant really think of any games with true circles and arcs.

I know nothing about computer programming soo

4/5/2006 8:17:20 PM

pixels aren't round

4/5/2006 8:28:42 PM

because back in the original days of 3d, a decision had to be made whether to do nurbs based rendering or polygon based rendering.

At the time, there were a lot of "tricks" that could be used to massively speed up polygon rendering, while nurbs surfaces were really really taxing on the first generations of hardware.

These days, if any company would take the time to develop a decent realtime nurbs engine, it would be comparable performance-wise to the high poly games we are now seeing, but would for the most part be MUCH more realistic looking. I have no clue why no development house hasn't jumped on this in the past couple of years, because there isn't really any technological reason not to anymore.

This is all based on superficial knowledge and reading on the subject back during the ATI smoothvision days (which took poly models and sort of converted them on the fly to nurbs). But I haven't seen any technical reasons against it.

4/5/2006 8:52:11 PM

I remember around the time that Quake III came out there was a lot of talk about moving to splines or whatever for stuff in 3D.

Of course, Carmack also said that within five years everything would be done with voxels and the average game level would take 20-30 gB of storage.

4/5/2006 10:13:43 PM

asdf

[Edited on April 5, 2006 at 10:13 PM. Reason : asdf]

4/5/2006 10:13:49 PM

carmack can say whatever the goddamn hell he wants

4/5/2006 10:35:36 PM

what the fuck is a nurb?

4/5/2006 10:52:11 PM

^ n00bz retarded cousin?

4/5/2006 10:56:22 PM

omg adrian carmack's name was all over wolfenstein

-ZiP!-

4/5/2006 10:56:33 PM

^^^ a NURB is a non-uniform, rational B-spline


cmon genius.

4/5/2006 11:11:22 PM

NURBS are slow. Using them in games doesn't have a good cost-benefit. It's pointless to spend time trying to draw perfect circular edges when antialiasing techniques work alright, and you can devote your time to other things such as testing, developing gameplay, network performance, etc.


[Edited on April 5, 2006 at 11:23 PM. Reason : fsd]

4/5/2006 11:20:08 PM

Yea, I love how neon thinks he knows more than the entire gaming industry.

4/5/2006 11:27:50 PM

he knows more than every industry

4/5/2006 11:30:51 PM

surface based models are expensive for real-time graphics, especially in situations where high poly count environments are likely to be encountered ... its about performance - theres a reason the game industry doesn't really touch it

4/6/2006 12:01:36 AM

its a lot easier to do math on polygons that can broken down into triangles and really simple low-level functions versus nurbs.

4/6/2006 12:38:51 AM

I don't agree with Noen, and here is why.

Quote :

"but would for the most part be MUCH more realistic looking"

No. Textures make the world look more realistic. Not a higher polygon count, or very slightly smoother curves.

Quote :

"I have no clue why no development house hasn't jumped on this in the past couple of years, because there isn't really any technological reason not to anymore."

Uh, how about the fact that it would be unrealistic to do in a realtime game engine? Not only have you failed to realize how many more calculations you would be doing from simply the barest of bone rendering. But you seem to have completely forgotten the use of new technologies. Dynamic lighting, shaders, parallax mapping. All of which would compount the needed calculations. Oh, and lets not even consider the possibility of ray casting lights if we switch to NURBs. Hey Guys, who wants to calculate the reflected ray, AND the b-curve it bounces off at. oh shit, wouldn't that be a pain in the ass. So shadow calculations would be harder as well.

Lets not forget, animations. Something tells me your b-curves wont behave very nicely any more when you connect them to a moving skeletal structure. But...that's just a guess from experiences with moving b-curves, connected to other b-curves.

But, assume none of that is a problem. Except for the known fact that yes, it will take longer to render. The solution is of course, Parallax mapping. If I take a high detailed model with millions of polygons and create a normal mapping texture based on that number of polygons, I can reduce the number of polygons to a couple thousand, and when I need to render the model, use the normal map texture as the normals to calculate the light from. Allowing me to create the appearence of the same million polygon model, but only using a few thousand polygons, or however many you need to at least get a good outline of the model so it does not appear rough around the silhouette.

There's no reason to switch to NURBs based modeling, when textures are finite in detail. I could care less how good of a curve it gave you, the fact that I can run up to objects and see very quickly the detail drop off as a run over and pickup the object on the ground, and see the poor texture. The fact that it's curvey wont even occur to me.

And there you have it.

Oh and one more I almost forgot. Realistic physics calculations. Have fun calculating that shit beyond the bounding box.

If you can solve the problem using more memory and less processing power. You 99.9% of the time always choose memory. It's cheap, and there's plenty of it in the world. Hence why the idea of using a normal mapped texture is so lovely.

[Edited on April 6, 2006 at 1:21 AM. Reason : y]

4/6/2006 1:10:59 AM

Look shitstains,

Ever heard of Shiny entertainment? Messiah, Enter the Matrix, Et al? Well they all run on nurbs models, which are tessellated out to the rendering API.

The reason no one uses Nurbs, is because neither DirectX or OpenGL support it. It's hard enough to get a new 3d engine on the market, much less write the API for the engine too.

Say what you want, but it's absolutely within reason for current gen cards to be able to render nurb models at a good, playable clip. As poly models go higher and higher counts, Nurbs become more efficient in pretty much every aspect.

^to reply to this.

You are comparing Apples to Oranges. It makes no sense whatsoever to take POLYGON specific rendering, lighting and calculation techniques and to say "omg this will be stupid on Nurbs". No shit sherlock.

Maybe you dont realize this, but you can do full skeletal animation with nurbs models in Maya and Max.

I'm not arguing whether one is "better" than the other, or which would would be faster. I gave the guy a snyopsis of the fucking history behind why we use Polygon based rendering API's for games.

[Edited on April 6, 2006 at 1:44 AM. Reason : .]

4/6/2006 1:20:48 AM

But they aren't going higher and higher, did you even fucking read my post. Most companies are switching to parallax mapping. Unreal 3 is using it in a big way. It's the next big thing now that people have vid cards with 512mb of texture memory, and 1gig+ of system memory.

Just because those companies use NURBs for the models before they tessilated them into polygon models means what? That they perfered using a simpler modeling format before taking it into a 'rasterized' version? Wtf does that prove. not a damn thing.

The reason you would do you models using NURBs on the modelers side is so, if you were developing a game for deployment over many platforms with ranging degree in graphics power, it would allow you to re-tessilate the orgional model for the next system without having to go back and manually add in more detail, or any at all.

[Edited on April 6, 2006 at 1:35 AM. Reason : bonus info.]

4/6/2006 1:25:12 AM

Whoops took too long to reply.

Dude. I'm not arguing the fucking state of the industry. All I'm trying to get across is that realtime NURBS/B-Spline rendering IS POSSIBLE.

No one is going to do it, because it would take a redonkulous amount of time to develop, when the payoff would be a marginal difference at best.

Jesus christ, way to go off on a tangent there.

4/6/2006 1:46:35 AM

OH SNAP!

4/6/2006 1:46:43 AM

I answered his question in three words. this dick wagging contest makes this shit sound complicated. pixels are fucking square (or rectangular if you want to be an asshat). you can't make true circles out of squares. end of answer. this guy didn't need a doctoral thesis on nurbs vs polygons.

4/6/2006 1:50:10 AM

Re-reading my original reply to this thread, I need to concede the reason this hasn't been done is due to retarded cost to payoff.

^He wasnt asking about that, but if you wanted to you could roll out an ol laser game that actually drew the objects to a screen, then there wouldnt be any pixels

[Edited on April 6, 2006 at 1:53 AM. Reason : .]

4/6/2006 1:52:05 AM

You're fucking stupid. He wasn't talking about anti-aliasing, or a lack thereof.

4/6/2006 1:52:52 AM

Quote :

"but if you wanted to you could roll out an ol laser game that actually drew the objects to a screen, then there wouldnt be any pixels "

true, but the Smithsonian would probably be pissed at you for disturbing their artifacts.

4/6/2006 1:59:38 AM

wouldn't matter if pixels were round. Then you would have gaps in the screen. Your 3 word explaination isn't as cool as my dick waving

4/6/2006 2:00:34 AM

^Are you takin Young's class this semester?

4/6/2006 2:51:17 AM

Already took it, both.

4/6/2006 2:52:10 AM

Young's class is for PUSSIES

4/6/2006 3:10:12 AM

looks like somebody failed after all.

4/6/2006 3:17:01 AM

please. Everyone knows that young is the biggest pushover of all the graphics teachers

4/6/2006 3:17:58 AM

ah, word. Did you do the co-op project with ID?

4/6/2006 3:30:47 AM

Yup. Were you in my class or something?

4/6/2006 3:33:03 AM

Quote :

"please. Everyone knows that young is the biggest pushover of all the graphics teachers"

Young doesn't teach graphics.

4/6/2006 3:37:42 AM

With a high enough polygon count and/or in addition to techniques like FSAA, this will cease to be an issue... of course, games will need to (and have started to, apparently) utilize models that can provide a variable polygon count to the rendering hardware based on its realistic polygon throughput to take full advantage of multiple generations of rendering architecture

It is a reality that many modern hardware acceleration techniques are simply not applicable to surfaces other than triangles, but considering the supposed current use of non-linear models and the blistering pace of graphics' product cycles it seems that the solution is already in the works anyway

Watch and wait, or something

4/6/2006 3:39:57 AM

This thread needs a healthy dose of Jaeger's input.

too bad he's in Prague with his snorkel and barmaid

4/6/2006 10:14:14 AM

Quote :

"please. Everyone knows that young is the biggest pushover of all the graphics teachers"

hahaha

young teaching graphics??

he couldnt even explain how to mod unreal, much less teach grahpic theory and application.

4/6/2006 11:02:52 AM

Quote :

"pixels aren't round"

hahahahahahahahahahahaha

4/6/2006 11:50:04 AM

Quote :

"Yup. Were you in my class or something? "

Nah, I'm in the studio next door though. What was yall's project?

4/6/2006 12:48:56 PM

Quote :

"young teaching graphics??"

I concur that Young is a goober.

Quote :

"If you can solve the problem using more memory and less processing power. You 99.9% of the time always choose memory. It's cheap, and there's plenty of it in the world. Hence why the idea of using a normal mapped texture is so lovely."

Not true on consoles. On PC, maybe; but if you think about consoles, which is where most of the industry is devoting their time, memory is a HUGE factor. Memory is at a premium on those things, so much so that you have to go through every level and optimize it so that it fits in the limited memory on those systems. Some engines dropped support for spherical harmonic maps, along with other rendering bells and whistles, for this very reason - the cost / benefit ratio of memory / performance just wasn't worth it for this generation. It is such a big factor that you have to develop custom software to stream stuff in from the DVD just to get only sections of a full level into console memory.

I'm not sure many people appreciate the engineering effort that goes into these games. You have to account for almost every byte of memory and be careful of how your data gets burned to the DVD to optimize seek time -- even duplicating data on the DVD in most cases for the sake of optical media bandwidth. On a tangent here, but don't even get me started on the cache performance of the in-order processors of the next generation consoles. Just think of all the branching and conditionals that happen in a typical game loop with AI and physics and what not, then realize that on next generation console processors you're going to genereate a cache miss a HUGE percentage of the time through that same code that performs "well" on modern out-of-order desktop CPUs. Insane.

Quote :

"Say what you want, but it's absolutely within reason for current gen cards to be able to render nurb models at a good, playable clip."

Not really. The model would have to be tesselated down to triangles before the hardware acceleration could really do anything with it. However, rendering is a task that can be split amongst multiple threads of execution, and eventually good general purpose CPUs will out-win GPUs in terms of raw rendering performance (ie, next-next-generation, when you have 10s of CPU cores).

Quote :

"This is all based on superficial knowledge and reading on the subject back during the ATI smoothvision days (which took poly models and sort of converted them on the fly to nurbs)."

I think you actually mean those N-patches, which was ATI's TRUFORM stuff. SmoothVision was just a fancy marketting term for ATI's MultiSampling FSAA technology. TRUFORM never worked out so well for them.

Quote :

"too bad he's in Prague with his snorkel and barmaid"

Fuck you, Tadd - you're just jealous! She's also a model, donchaknow.

4/6/2006 1:43:54 PM

who the fuck is Tadd??

4/6/2006 1:59:50 PM

^ Whatever, he knows who he is.

4/6/2006 2:00:37 PM

Quote :

"Not true on consoles. "

Very true. I think the PS2 has like 32mb of video memory. Some piss small amount. I remember the devs talking about how they had to use crazy trickery to get around that problem.

So yeah, consoles are a unique breed. twas why I left the .1%. But I was speaking more to PCs and the next gen consoles when making the future choice on which path to choose. On the nextgen boxes, space should not be an issue as it has been in the past.

With NURBs, the thing is that they only save you space when describing simple objects. Given the constantly changing shape of objects it would take just as many points as we current use to describe an object..maybe slightly fewer in certain situations. but each point requires, more information to record the b-curve at that point. But you will have smoother figures.

But, you are still going to want bump/normal/parallax mapping. Just because you need to be able to represent a "texture" to the surface. In either case, you are going to be employing the use of parallax mapping if you want good models. So you may as well let it take care of the problem you are having with curved surfaces, and continue to use polys and keep as many calculations as you can, as fast as possible.

Quote :

"However, rendering is a task that can be split amongst multiple threads of execution"

which makes shaders a damn nightmare for graphics board designers

[Edited on April 6, 2006 at 2:17 PM. Reason : hehe]

4/6/2006 2:13:43 PM

Quote :

" think you actually mean those N-patches, which was ATI's TRUFORM stuff. SmoothVision was just a fancy marketting term for ATI's MultiSampling FSAA technology. TRUFORM never worked out so well for them. "

Yea it was TRUFORM, I missed my edit window on the post.

Quote :

"Not really. The model would have to be tesselated down to triangles before the hardware acceleration could really do anything with it. "

Since all these new videocards are supposed to have "programmable gpu's" shouldn't they be able to be useful for splines as well? Or are they, like the API's, just completely "programmable" as long as you want to do poly rendering?

4/6/2006 3:30:31 PM

The programmable gpu is just for shaders. The data structures are still the same, it just allows you to do some post processing on the image. You just get the vectors and normals as input into a method. From there it's up to you to give it back colors or changes to the vectors, or to do color modulation to the texture.

4/6/2006 3:42:42 PM

What part of pixel and vertex shaders (the programmability of next-gen GPUs) might make one assume it applies to polygons?

Ahh, yes...

Quote :

"Vertex, noun the point in which the sides of the angle meet."

The GPU processes triangles, textures, and pixels... you can't fit a NURBS peg in a triangular hole, even if you really think the architecture should be able to do it

4/6/2006 3:43:55 PM

That's what I figured, but I thought it worth asking.

Zero: Which group were you in?

4/6/2006 5:33:34 PM

Angrymob (the one that did Wideasleep)

4/6/2006 10:26:50 PM

Ah so you worked with my boy Alfie, that was the only group that made anything worthwhile (the sidescrolling ninja game was a good idea though)

4/6/2006 10:40:16 PM

Yup. Alfie and Brian. Did you go to the presentation?

4/6/2006 10:44:59 PM

Quote :

"Are you kidding me? I thought it was great! Your argument for the inverse ratio of capitalization to debt shifting 3D modelling to nurbs was genius. Now if we could only get congress the entire gaming industry not to be so short sighted."

[Edited on April 6, 2006 at 10:47 PM. Reason : .]

4/6/2006 10:46:09 PM