Quads vs Triangles workflow


As a teacher and artist, I’ve often practiced, reviewed, and taught the principles of good / clean geometry. Naturally, that means pushing a strictly quad-based workflow. However, I understand that game engines tend to prefer triangulated models for collision detection and render performance. So here’s the question, why push the use of quads in the first place?

Now, I’m not talking about N-gons. Those are a different beast. We all know that N-gons cause problems with animation and rendering. Moving on…

Getting back to the question at hand, let me reinforce a bit. If the GPU renders triangles faster than a quad, and a quad is really just two triangles with a hidden edge, and the game engine prefers tris, then why not just use them?

I was reading an article on Plural Sight https://www.pluralsight.com/blog/film-games/ngons-triangles-bad that stated triangles and n-gons cause issues with animation and rendering. I know this is true with n-gons (I’ve experienced it first-hand) but how could it be true about triangles? Once again, a quad is just two triangles with a hidden edge, which means triangles are always present even if your workflow is in quads. So how could a triangle cause rendering or even animation issues?

I personally have never had a such a problem with triangles, has anyone else?


Couple of things to consider off the top of my head…

  • clarity of loops in relation to anatomy
  • harder to manage poles with tris
  • normal computations are pre computed rather than post tessellation
  • solver preference for mesh deformation. Eg cloth cross links
  • quads -> tris = easy. Tris -> quads not so much
  • some important common modeling tools depend on quads (eg. select edgeloop)
  • A big one - Subdivision! ( give it a try…yuck)

probaby a bunch of other things I’m missing and I’m rusty with my modelling… but yeah . You can use triangles (seen some great use of them with intent even) but knowing where you mesh is going and it’s purpose - is key.


revilo3d has named most of the points already. Which is mainly about modeling. The two most important biggies for me: No edgeloops means you cannot use any edgeloop tools. And SDS makes poles with tris. A quad mesh is simply in general easier and faster to make.

You have also to distinguish between modeling and the end result, the use case. Modeling for game content also mainly happens with quads. But they get triangulated for the game then most of the times. Nowadays game engines can handle quads, but those importers sorts most probably the underlying triangulation of quads in a different way than your modeling software. And then it can happen that you have ingame unwanted edge directions and distortions at the mesh. Quads can work. But there is no guarantee that the edges points into the wanted directions.

And last but not least, tris in games has also historical reasons. Most game engines nowadays also supports quads. But older game engines and graphics cards could just handle tris.


Quads are good a good choice for editing, animation, predictable smoothing, and easier to lay out 2D UV’s for editing in 2D programs

If you’re you don’t have any of those requirements, then triangles are fine and can be more efficient, especially for high-res meshes for offline rendering that either already have UV’s or that don’t require UV’s - such as the case of using procedural volumetric 3D textures

For static meshes, IMO decimated meshes are much more efficient with polycounts than quads. Like if you have some scan data that you just need to drop in your scene and be done with. You could go to the trouble of retopologizing it, but if it’s not going to add any value or capability that makes a difference, why bother?


Simple tests with basic geometry, using 2-3 bones or biped/cat will very fast show what the all fuss about.
I really encourage anyone studying topology for character or sub-d modeling deforming meshes (and that’s quite a different thing than static meshes) try very basic rigging with skinning, and see for yourself, what happens when you start bending knees, fingers, start twisting the body. You can use very simple meshes to test it. Just try it with different loop direction and uniform\long quads.
It will take about a week to learn basic skinning and basic rigging. Any starter videos from youtube will do.