Highpoly "discussion", formerly known as MC0003

Optimus Prime · Post by **Optimus Prime** » 25 Oct 2006, 16:30

the main point why i began to ask "why not make it a more polys per model competition" was: a maximum of 1000 tris is good for uncomplicated models like simple tanks or simple aircrafts, but if you want to make some nice vehicles with a lot of wheels or other parts like different weapons, you cant do them or at least not very good. You are forced to cut details to reach the <1000 tris limit. So all i wanted is the opportunity to make vehicles that are more than a tank with 2 sausages as tracks, a car with 4 wheels or a simple hovercraft.
A simple 8 sided wheel needs 2*12 + 2*8 tris = 40 tris per wheel, so 160 only for the wheels (and 8 sides are very low in my oppinion and dont look round enough).
I think all my war evo models dont have more than 10% of wasted polygons and not one of them has less than 1.5k polygons, so i would at least double or tripple the ammount of polys for a competition. Think of krogoth sized units.
I never said or wanted to waste polygons to increase the count.

aGorm · Post by **aGorm** » 25 Oct 2006, 17:42

You started to come across like that though, which is mainly because you argued against people that said "I can make what I want it 1000 polies fine". IF you can make your moidle in under 1000 polies, then there is no reason to use more. People took you to be saying "wast polies!1!!1" because you kept saying you should use more polies, when they didn't want or need to on the models they made.

Can a mod lock this or somthing... I mean, its realy pointless argueing over somthing so trivial.

aGorm

rattle · Post by **rattle** » 25 Oct 2006, 18:00

^
Look at your model, the tire treads especially. You are wasting polygons. Everything that can be done by texture instead of using extra polygons are wasted polygons. I'm not saying make your tires 8 sided cylinders, it's about the extruded things... it's just one example.

Ah what ever I have enough of this, other things need more attention.

Optimus Prime · Post by **Optimus Prime** » 25 Oct 2006, 18:02

i said i didnt wasted more than 10%! I didnt said i have no wasted polygons.
The simple reason is, that the models arent that detailed and war evo is made for very few units, so there is no need to reduce them.

Peet · Post by **Peet** » 25 Oct 2006, 20:45

Except that it runs slower than LARGE GAMES on XTA with the same number of players.

KDR_11k · Post by **KDR_11k** » 25 Oct 2006, 21:00

These units have to fit into different kinds of mods which means they should be at a number that works well even in large amounts. It's nice that your mods can use the higher end of the poly range but mods like yours are the exception rather than the rule.

And 8 sides are plenty for a wheel in an RTS, you can use even less without anyone noticing.

SwiftSpear · Post by **SwiftSpear** » 26 Oct 2006, 03:56

I can't tell why there is an argument in this thread... As far as I can tell people are arguing different semantic points... but just to see if there is someone crazy I'm going to ask a few questions and see how people argue.

Is it appropriate to not optimize your models for in game?

When referring to polycount on models, does high poly count = a better model?

Does texturing matter to how a model looks in game?

Everyone arguing please answer so I can figure out who is right and who is wrong, because as far as I am concerned there are very simple objective answers to each of these questions.

Snipawolf · Post by **Snipawolf** » 26 Oct 2006, 03:59

Is it appropriate to not optimize your models for in game?

Of course not, especially if you are trying to get a job. Details are better done with textures in plenty of cases.. Small objects (The wheel extrusions) are pointless and not needed.

When referring to polycount on models, does high poly count = a better model?

Of course not, the lower you can go while retaining details is best. It's more like the other way around.

Does texturing matter to how a model looks in game?

No, but many simple things like bolts, nuts, edges, grooves and many other things are easily done with textures.

I hope those were right..

Argh · Post by **Argh** » 26 Oct 2006, 04:18

Gonna paraphrase the first question to this, "Should you optimize your models for good performance?"

Yes. Absolutely. And, is every model I make a paragon of the above? No. My first models for NanoBlobs are actually poorly optimized, and could have been done more professionally.

When referring to polycount on models, does high poly count = a better model?

No. In professional circles, one talks about the specification to be met. For a typical Spring mod, the reasonable specification for any game model is under 2000, and under 1000 is ideal, in my opinion. Do I meet that spec? Yeah, but not always as well as I should. Usually, I have been more concerned with getting things done, or tricks like the Knight's treads, than I have been with sheer technical perfection. Still, I would argue that even in a mod like War Evo, one should not see models with polycounts over 4000 triangles. Because, quite frankly, you can put really insane amounts of detail into something with that spec.

Does texturing matter to how a model looks in game?

The basic skin, plus whatever fancy upgrades are available (everything from glowmaps to normalmaps, depending on the game engine) is, in my opinion, about 50% of how "cool" the final model is. About 10% of the "cool factor" is the geometry, and about 40% of it is the animations.

I tend to stress skinning very heavily to newbie modelers, because it's often seen as an afterthought by those who are learning the art of CG. Folks... it's probably the most important part of the art. Sure, modeling is hard... but skinning is usually much harder! Learning it as part of your workflow as early as possible is the best way to reach a high level of skill. I seriously doubt that any future technology is going to change this- it will always be faster to paint a lot've details than to simulate it with other methods... so I think that people should learn it well, and learn it early. Remember... nobody hires guys who can't skin. If you want to model professionally... it's just that simple. And if you want your models to look hot in Spring... you're going to have to learn the skillset.

This is all pretty subjective, and really depends on the individual work, of course.

For example, the Dreadnoughts from Dawn of War are not really detailed models with super-awesome textures, if you pause it and look really closely. They have blurry, low-res textures that mainly look good because you can't see them very well while the units are in motion. What really makes them work so well, in my opinion, is that the designers were aware of this, and used constant animation to distract the viewer. There is no realistic reason why a Dreadnought, which is a machine, should be in motion when not moving, for example... but the designers made them seem to be alive with looping animations while still. This really works wonderfully, and the Dreadnoughts all have a very distinctive feeling of character and life that really makes them work well.

Optimus Prime · Post by **Optimus Prime** » 26 Oct 2006, 04:40

SwiftSpear wrote: 1. Is it appropriate to not optimize your models for in game?

2. When referring to polycount on models, does high poly count = a better model?

3. Does texturing matter to how a model looks in game?

1. there ARE situations, where it is NOT necessary to optimize your model, just because you know, that the details are that low, that no one will ever notice a difference in performance, while optimizing needs time to rework your model.
Just think of a millionair who wants to pay 97 cent for a bottle of water. Wouldnt he give 1 dollar and go instead of waiting for the rest money? I would.

2. No. BUT you can make more beautyful models with higher poly counts. So its like this
higher poly count >= better model (if good means beautyful)
higher poly count < better model (if good means performance)

3. Textures are very important. Its the skin and their clothes together, but in my eyes your texture must be as good as the model, so its useless, if you have a super texture on a very bad model and even the best model is wasted, if your texture looks bad.

I think the problem of this discussion is that people began to talk about different things. At the beginning it was just about polygons, if you can make models better with higher poly counts and than people throw textures, animations and so on into the discussion. Sure they all are connected together.
But its like if you want to know, if a better motor means a faster car and than people begin to tell you that better wheels mean that you will be faster. Both points are right, but the second one is not the answer of the question.
(OK, now the discussion is about more than just polygons, so textures are welcome too)

Argh · Post by **Argh** » 26 Oct 2006, 05:16

there ARE situations, where it is NOT necessary to optimize your model, just because you know, that the details are that low, that no one will ever notice a difference in performance

That is simply not true. You should always be keeping a stern eye on polycount. A model should always have the minimum polycount needed to meet design requirements.

And you should always be aware that your initial polycount is actually going to be multiplied during the actual rasterisation process.

Put another way... if your mod has only 4 models onscreen at once, EVER, then the maximum polycount, per model, should be 15,000. Why that ceiling?

15K, shading step: 15K
15K, smoothing step: 15-30K, depending on a host of factors.
15K, glowmap: 15K
15K, reflectionmap: 15-45K, because of tesselation
15K, shadowmap: 15K
15K, transparency: 15K

Grand total, triangles per pass: 120,000 triangles.

So, for 4 models, we have a grand total of 480,000 triangles per rendering pass... plus the mesh for the map, which if it's SM3, may also include a shader pass, tesselation due to water shaders (that's a huge number of triangles) and other crap going on.

Now, keep going... let's say our FPS is 60. 60 rendering passes need to be made. Let's be really nice, and rate the map mesh at 5K triangles, no shader passes, no grass, no trees, no smoke.

485,000 * 60 = 29,100,000!!! That is how many triangles are being shoved through the GPU, plus a huge number of shader calculations, textures, normal coordinates, and a bunch of other data!

To put this another way... the average polycount in NanoBlobs is about 1500 triangles. Typically during play, you'll see about 20 of them onscreen, max, if combat is really heavy. That is about 30K triangles, or half of the load of your 15K super-robots. And guess what? They cause massive, chronic lag for most users with typical systems, if they have any of the bells and whistles on. On my rig, which has a pretty decent GPU (7800OC), I get about 23FPS when viewing full-on scenes of massive carnage, with shadows, reflections, and other stuff on. That's pretty good! And I only got to that point in the first place after months of work optimizing the mod for speed, because people complained so much about how bad the performance was

you can make more beautyful models with higher poly counts.

Yes, but it's definately a matter of "by how much, for how much cost". A 10% increase in polycount, from 1000 to 1100, can give you a much bigger gain, in terms of detail, than an increase from 10K to 11K can. 100 more triangles, at the low end, means that you can add a few new boxes, trimids, or simple columns to add a great deal to the detail and feel of the piece. At 10K to 11K, we're talking about very minor feature greebling.

In short... if you want to push out more polygons... build something that is as low-poly as possible, then add greebles, one small bit at a time. Weigh the artistic merits of each new shape carefully against the overall polycount. When you are near the spec. for the game engine... you need to make hard decisions about this- either lose some detail elsewhere, or not add any further. You should not just be, "but it needs this to be complete" at that point. If you didn't do drawings before you modeled your piece, so that you know what it needs to look like before you started, then you've just learned the hard way why almost everybody who knows what they're doing does so

Half the reason newbie modelers suck is because they just sit down and start modeling, instead of doing drawings until they have a hot design... then they end up with an unworkable mess

in my eyes your texture must be as good as the model

Bullcrap. Go look at models from old games. They are extremely low-poly, yet they worked because of the skins.

When you actually know how to skin, you will not ever say that again. You simply don't know what you're talking about.

SwiftSpear · Post by **SwiftSpear** » 26 Oct 2006, 06:36

I'm a bit split, because currently in the video game industry it's standard to make the first pass model anywhere from 1-5 million polies and then remake/optimize a second pass model. The first model is used to derive a UV map and the second model is textured, UV applied, and put in game.

I don't think it's ever ok not to optimize, the low poly model should be as low poly as you can get away with. That isn't to say you shouldn't be modeling high poly, and in the case of your high poly model optimizing doesn't matter, but you're just ripping off yourself and the gamer if you don't optimize a low poly model. The UT2007 models are ~5000 polies, yet the detail is stunning, they are optimized as hell if you acctually look at them, they just look really good because of excellent texturing and effective normal maps.

[edit] I'm with argh 100% if anything model geometry is only 10 percent of the acctual model quality. You can make UGLY geometrys look really good with good texturing and animation.

rattle · Post by **rattle** » 26 Oct 2006, 12:11

And the proof to that has already been posted some pages back: Star Wars Mon Calamari infantry. Extremely low polygon model but immense amount of detail thanks to their awesome skinner.

Optimus Prime · Post by **Optimus Prime** » 26 Oct 2006, 12:13

i m still not happy with that.
Put an ut2007 texture on a quake 1 model and in my eyes it still looks not good. Its just because you can see that all the texture details are made and only made by textures while if you have an ut2007 model, you dont know this because all texture details have simple polygon forms.

And to the first point, argh take instead of 15k polys 1.5k polys. So your for 4 models you would have 48k tris per scene. Even a geforce 2 can handle that with ease. So why should i optimize the model from 1.5k to 1.4k if no one would ever notice a difference in game? Only because i can say later "ohh all my models are so optimized"?

And the proof to that has already been posted some pages back: Star Wars Mon Calamari infantry. Extremely low polygon model but immense amount of detail thanks to their awesome skinner.

and as i said, they look good for spring, but for an example to this discussion, they are not. Perhaps we speak on different levels - i m discussing more than we can see in spring, i want to speak about poly count in general (so take modern games as a limit) and for a company of heros model, it would look like the low detail version if you put your settings on lowest.

imbaczek · Post by **imbaczek** » 26 Oct 2006, 12:20

You still fail to understand one thing: looking good IS NOT what games are about, even though it's a huuuuge part of them. Nobody will care that your game looks great if you need a freaking renderfarm to run it.

Argh · Post by **Argh** » 26 Oct 2006, 12:22

Because when you go from 4 models to 40, it rapidly adds up. That is my point. If you don't need those 100 tris, they should get eliminated, if they aren't absolutely necessary for the art.

Swift's point about where modern games are going is true, insofar as where the latest techniques are going. Problem is... Spring doesn't support those techniques yet. OSRTS may do so, but even then... for older systems that don't support the latest shader models, this stuff is basically useless, and while I have seen normal maps in the latest RTS games, and they really do add a bit of nice detail, I'd be the first to say that they're not a cure-all. Just like everything else, normal maps come with a cost. But at least they provide a good end-run around having to do 5 LODs for one model

Snipawolf · Post by **Snipawolf** » 26 Oct 2006, 13:01

Argh, never say modelling is hard.. Even slightly...

I reduced a buggy from 700 tris to 300, its really easy to do it.. Then pretty easy to go back and fix mistakes..

Going to the company of heroes subject, those guys look awesome.. Why?

Not models, my friend, textures.. It's the only thing on my demo I can turn up without slaughtering my FPS and it looks great..

Argh, you said you would be finished by like.. Yesterday

Argh · Post by **Argh** » 26 Oct 2006, 13:06

<shrug> if you haven't noticed, I've been a bit busy. Happens. And I still haven't quite decided how I'm going to a certain thing with the unit, which is why I've been procrastinating

Snipawolf · Post by **Snipawolf** » 26 Oct 2006, 13:11

Well, take a couple of design guesses and try some stuff, never know what works..

Argh · Post by **Argh** » 26 Oct 2006, 13:17

It's not that simple. It involves a fair number of moving parts, and I have to pre-plan some of it rather carefully. You'll see what I mean when I present it.

Spring RTS Engine

Highpoly "discussion", formerly known as MC0003