There is an age-old debate when it comes to polygon modelling.  Lots of folks have strong opionions on what good topology is, and they judge the quality of a model based on the use of quads vs. ngons (polygons with more than 4 sides), or tris.

I hear artists insist on always use quads, and nothing else, when modelling.  As someone who makes models for a living, often under unbelievable time constraints and pressure by clients, my opinion is that Ngons are not only acceptable on a model in many cases, but often a good way to efficiently reduce production time, and save on resources.  When it comes to static, flat (planar) surfaces, Ngons are perfectly fine to use, and sometimes strategically intelligent!  By definition, triangles are planar, so it makes perfect sense to incorporate them into planar surfaces, and could possibly save you some rendertime in larger scenes, as your engine doesn't have to break down the quads or ngons into triangles.   Using quads for flat, planar surface space on models is just a waste of resources, and time.  I know, it looks impressive in your wireframe renders if everything consists of uniform, quad faces, but it isn't always the smartest modelling technique.  When it comes to curved surfaces, or organic models, quads or tris are better solutions, because Ngons can't flow well on curved surfaces, but that's the only situation where I'd advise to stay away from ngons at all cost.

When it comes to RT processing efficiency, triangles are the best, and only way to go.  Triangles rasterize quickly, and make it really easy for the render engine to process.  If you're goal is output to RT (Real-Time) engines, then using triangles is easier and faster than dealing with quads, which is why game engines use triangulated faces.

The only time you should really be concerned about keeping things all quads is when modelling organic objects, like humans, or in areas of the topology that will deform in animation.  Quads allows you to easily access the flow of the mesh topology, through edge loops, and therefore makes it much better to control the overall smoothness and curviture of the mesh when you apply Catmull-Clark subdivision.

So should you, as a modeller, follow the mantra that you should always model in "all quads" all the time?  No.  It's not efficient, nor intelligent, to just mindlessly keep all your models quads at all times. Should you avoid Ngons like they are the devil himself?  Again, NO.  Ngons can be used in many cases to speed up workflow, and reduce overhead in a model.  Just know when and where they can be used successfully, and you are fine.  Remember, if it looks good rendered, then chances are, it's good.  Are tris ok to leave in a model?  In my opinion, by all means, yes.  Everything, in the end, gets triangulated upon render.  Quads are just two triangles together.

Bottom line, when it comes to building models, don't just follow a senseless motto "all quads".  Know when and where it's ok to use tris and ngons.  Unless you're in a contest for the prettiest wireframe render, it's just intelligent modelling strategy.

Thoughts?