Why to choose Faces or Tris for modeling? Pros vs Cons?

DefoNotMike

Hello there, I had a question about Faces and Tris in 3D Modeling. I understand the difference between them in the sense of what they are.
What i am confused about is:

-Whats the difference to using them?

-Are there advantages/disadvantages to using one over the other?

-When modeling should I be watching my Face count or my Tri count?

-Is one better for optimization in an engine than the other one?

Thanks again for taking the time to review my question friends. Cheers.

CarlCraft

What do you believe the difference between a face and a tri is? A tri is a triangular (3 vertices) face. Do you mean tri and quad (4 vertices)?

EDIT (to be more helpful and not only seem smug): 

ZacD

-Whats the difference to using them? 

Quads are generally more flexible and are what modeling tools are designed around. With quads you can easily add or remove edge loops, which isn't possible with tris.

-Are there advantages/disadvantages to using one over the other?

Generally, you should be using both. Mostly quads, but triangles where necessary. 

-When modeling should I be watching my Face count or my Tri count?

Don't matter much as long as you are consistent.

-Is one better for optimization in an engine than the other one?

By the time the GPU renders a 3d object, everything is converted to tris. You should convert the model to tris before baking the normal map. But keep a quad version of the model in case you need to change anything. 

ActionDawg

CarlCraft said:

What do you believe the difference between a face and a tri is? A tri is a triangular (3 vertices) face. Do you mean tri and quad (4 vertices)?

EDIT (to be more helpful and not only seem smug): 

say it with me yall:

ngons are fine. just another tool in the toolbox.

ZacD

Yup, often times n-gons smooth better than tris. 

danr

jesus. Burn that image with fire,  and never speak of it again

Mark Dygert

Booo hiss... (throws popcorn) that image is full of bad info. Not only does it offer bad info about the use of ngons, but it doesn't even define ngon correctly. That makes artists look stupid. That is a pentagon not a ngon, unless you're retarded and can't recognize a 5 sided polygon. 

There are specific names for polygons that have more than 5 sides:
5 pentagon
6 hexagon
7 heptagon
8 octagon
n = an indeterminate number, ngon.
How many gons?  I don't know and I don't care, there are too many to count. There are "n" number of gons. It is an ngon. Usually after 8 sides (octagon) the shape becomes so rounded it's hard to tell and people just give up and call them ngons.

-Whats the difference to using them? 

It's all polygons man... do whatever to get whatever.
Triangle modeling gets extremely tedious really fast on anything other than extremely low poly models. anything more than 500 tris is going to be a pain to tri-model. Quads let you work in edge loops and rings, which opens the door to a lot of techniques and tools.

-When modeling should I be watching my Face count or my Tri count?

Tri count. Faces will include ngons which will be counted at a single face, but really it could be a bunch of triangles.
Actually...  If you want to be exact, you should be paying attention to total vert count which also includes extra verts from UV seams and smoothing breaks. Ballparking with tris is probably fine for most objects in most game engines, just don't do anything stupid like facet a model or make every edge a seam and you should be fine. Just know that your model is a cloud of verts and a lot of operations run over verts, mostly transforms and shaders. Faces are just drawn between verts, almost as an after thought. 

-Is one better for optimization in an engine than the other one?

It all gets broken down into triangles. Non-triangle modeling is just for your modeling workflow.

sprunghunt

somedoggy said:

CarlCraft said:

What do you believe the difference between a face and a tri is? A tri is a triangular (3 vertices) face. Do you mean tri and quad (4 vertices)?

EDIT (to be more helpful and not only seem smug): 

say it with me yall:

ngons are fine. just another tool in the toolbox.

Many tools won't draw N-gons. 

For example Substance Designer will display n-gons as empty space. For any kind of real-time application n-gons are bad.

sprunghunt

DefoNotMike said:

-When modeling should I be watching my Face count or my Tri count?

You should watch your vertex count. Vertexes are what take up memory not tris. And the vertex count varies quite a bit by how many edge splits you have. A low vertex-to-poly ratio is as sign of a clean model.

Alex_J

I always avoided n-gons because that's what lots of people said to do. But when I thought about it, I didn't see any reason not to use for static meshes. Still, I avoided them because people higher than me said so. 

It's not a big deal to avoid using n-gons so I'll probably keep going that way out of habit, but some programs (like Zbrush I'm pretty sure), will complain if you import a model with n-gons. So what's that about?

Bruno Afonseca

DefoNotMike said:

Hello there, I had a question about Faces and Tris in 3D Modeling. I understand the difference between them in the sense of what they are.
What i am confused about is:

-Whats the difference to using them?

-Are there advantages/disadvantages to using one over the other?

-When modeling should I be watching my Face count or my Tri count?

-Is one better for optimization in an engine than the other one?

Thanks again for taking the time to review my question friends. Cheers.

- well placed triangles help you keep control the flow and silhouette of your mesh. USE BOTH!
- quads are easier to edit around, create loops and whatnot. USE BOTH!
- your in engine vertex count (don't trust max or maya). triangle count gives you an idea, poly count is useless
- everything becomes triangles in the end

reading material:
http://www.ericchadwick.com/examples/provost/byf1.html
http://www.ericchadwick.com/examples/provost/byf2.html
https://simonschreibt.de/gat/renderhell/
http://www.fragmentbuffer.com/gpu-performance-for-game-artists/
http://wiki.polycount.com/wiki/Limb_Topology

sprunghunt

BIGTIMEMASTER said:

I always avoided n-gons because that's what lots of people said to do. But when I thought about it, I didn't see any reason not to use for static meshes. Still, I avoided them because people higher than me said so. 

It's not a big deal to avoid using n-gons so I'll probably keep going that way out of habit, but some programs (like Zbrush I'm pretty sure), will complain if you import a model with n-gons. So what's that about?

Zbrush has a unusual 3d context. It's not displaying polygons like any other program. So it complains because of that. 

The problem with any polygon that isn't a tri is that you have to convert to a triangle to display it. This doesn't matter if the polygon is planar but if you have a non-planar polygon it can have several states depending on how it is triangulated. Here's some examples:



The two planes in the upper right are exactly the same number of vertexes but are triangulated in a different way. Below you can see a N-gon which is triangluated in two different ways. You can see that the wedge in the middle disappears in one example. 

Alex_J

So, to break this down very simply, no matter how many sides a face has, they always get digested into triangles. How you choose to work is totally about convenience, speed, efficiency, etc. Potential problems (shading/rendering) may occur if the computer has options in how it will divide quads/n-gons into triangles, and so in Zbrushes case the program demands that you define the triangles for it. 

This is correct?

musashidan

The on-card vert count is the true evaluation of the mesh. It is a calculation of physical verts plus UV/smoothing/Mat ID vert breaks. There are scripts to view the actual vert count, or in a program like Marmoset Toolbag this is already calculated and accounted for in the vertex count display.

I still can't believe how prevalent the N-gon myth is among newcomers to modeling(and many experienced modelers, it seems) The N-gon is like a poor bastard who was falsely accused of rape as a teenager, entirely cleared by the justice system, but whose life is utterly ruined by the accusation of 20 years ago.....

Using N-gons is vital to my hard-surface workflow. They allow me to model complex shapes very quickly. Then it's just a matter of adding a Turn to Poly modifier non-destructively to triangulate the mesh before export to bake/engine. Once you know how to place a few strategic edges to control the auto-triangulation and avoid extreme tris/shading artifacts then you're golden.

Even when sub-dividing the same applies as above. I use shitloads of N-gons. And again, once you know where to place a few edges at certain places on the topology to avoid artifacts, and a non-destructive chamfer/subdivide modifier, life is good. The sub-D mesh(in our case as game artists) is solely to bake from. If it looks good and smooths well.......then perfect.

All N-gons ultimately become quads when subdivided: 5x4=20/6x4=24/7x4=28.........and on and on.

Mark Dygert

sprunghunt said:

Many tools won't draw N-gons. 

For example Substance Designer will display n-gons as empty space. For any kind of real-time application n-gons are bad.

Yeah good point, its why most people triangulate their models before doing materials, especially if there is a normal map involved. If you aren't doing any modeling, you don't need the edge loops or rings and you want to make sure the tri-striping is constant between apps. Any difference will screw up your normal maps and give you seams.

A quick note about that, not all apps handle non-visible edges the same way, max lets you flip them any way you would like, other apps force non-visible edges to flow in a certain direction, others go the opposite. 

Alex_J

musashidan said:

I still can't believe how prevalent the N-gon myth is among newcomers...

Using N-gons is vital to my hard-surface workflow...

Well, I got started from pluralsight (digital tutors) and I think they tend to drastically simplify their beginner stuff. They do a lot of showing but not always explaining the why's, but I think that is for good reason. There is only so much information a newcomer can take in and digest. 

Case in point, all the stuff you've explained about your hard surface workflow -- I've been modeling all day every day for about 8 months now, and not much of that made sense to me. But I know enough that I could figure it out and make sense of it with a little research, but if you had tried to explain that to me 6 months ago it would be like trying to train a cat. For beginners, simple rules I suppose have a place just to get you going and avoiding things that could be bad habits if you don't understand how the background processes work.

musashidan

BIGTIMEMASTER said:

musashidan said:

I still can't believe how prevalent the N-gon myth is among newcomers...

Using N-gons is vital to my hard-surface workflow...

Case in point, all the stuff you've explained about your hard surface workflow -- I've been modeling all day every day for about 8 months now, and not much of that made sense to me. But I know enough that I could figure it out and make sense of it with a little research, but if you had tried to explain that to me 6 months ago it would be like trying to train a cat. For beginners, simple rules I suppose have a place just to get you going and avoiding things that could be bad habits if you don't understand how the background processes work.

It wasn't always like that. I learned the hard way. When I first started modeling many years ago subdivision modeling was almost like a religion. The days of stitching patches together in a science lab were behind the leaders of the community, Subdivisionmodeling.com was the temple where people worshiped and woe betide anyone who blasphemed against the almighty quad.

I spent years learning sub-d modeling for offline rendering/animation before I got into game art and laugh now at all those hours I spent re-routing edgeloops and stressing out over 6-valance poles and the fact that I had an extra vert that was not quad-compliant.

But it definitely had its merits as I obsessed about it that much that it gave me a very solid foundation and understanding of topology.

These days I have since become quad-agnostic and lay daily offerings at the shrine of N-gon.

Blouko

I've been told, 

That if you're looking forward to animate or use blend shapes on your character or model, it is better to use quads over n-gons or triangles. As the topology will move better if it's made up of quads, but that's mostly for characters.  

Also, I've been told that all-in-all, game engines triangulate your models on import. So, it's better to use quads, because the engine would automatically triangulate your mesh, which might lead to errors or weird topology if they are n-gons.

Although, I'm not completely sure of the information I've passed on, if I've made any mistake please point it out.

musashidan

Blouko said:

I've been told, 

That if you're looking forward to animate or use blend shapes on your character or model, it is better to use quads over n-gons or triangles. As the topology will move better if it's made up of quads, but that's mostly for characters.  

Also, I've been told that all-in-all, game engines triangulate your models on import. So, it's better to use quads, because the engine would automatically triangulate your mesh, which might lead to errors or weird topology if they are n-gons.

Yes, quads are easier to work with. Easier selection/control of edgeloops. UV unwrapping is easier. Rigging is a LOT easier. Meshes deform better(game characters are still ultimately triangles though) But all quads are 2xtriangles anyway. It's just that the edge that bisects the quad is a hidden edge.

Real-time engines do render triangles. Locking in your triangulation before baking and export to engine is standard. But this can be non-destructive with Max's Turn to poly modifier, for instance.

So yes, quads have their benefits for certain. But so do tris and n-gons. The point here is to understand and know why and when to use each, rather than blindly following a mantra. Perfect example featured in the image at the top of this thread. Nicely presented, but full of bullshit.