I just don't understand the point of sub-division for 3D models

Yes it results in a smoother looking model but then you have to retopologize it anyway thus making it no longer look smooth so what is the point? Why not just model it conventionally and save yourself the effort of having to go through retopology later? Why even use sub-division modeling at all if your intention is to use the model as a game asset? Even sculpting seems pointless to me for the same reason except as a means of creating normal details to be baked unto the actual model.

 

It is all for making the high resolution model to bake a normal map from. That's the only reason to do it. (unless you are making a high resolution model just for the heck of it)

 

Occasionally I use it to make a low-poly model from a ridiculously low-poly model. That is to say, you don't necessarily need to retopologize your model. Only if your model has an inefficiently high number of polygons.

 

Maybe this is a bit of a "back in the day" answer, but have you considered that there should be consistency between a model that appears in-engine and one that appears in CGI cutscenes, trailers or printed materials? I know polygon demos are today's default, but the fact we must still be aware of the amount of vertexes and lighting we put into a game asset means there might still be situations in which rendering a CGI movie would be preferable.

See: any Final Fantasy intro since the PS1 era.

 

And it's a workflow to separate design phases, with optimization phases. Ie we separate the concern so they don't interfere with each other. While designing, you should focus on making the model as good as possible, by any means necessary, good topology is a distraction. Once you finally have the correct result, well you can optimize the mesh for animation and polycount, knowing full well the topography won't change.

Subdivision is also useful to manage going from general to details, which is how you generally approach designing forms. It's a matter of common sense, general form have the biggest impact, locking them first is important, and it's easier to manipulate when there is a low count of primitive. Then you can fine tune by adding details. That's also how you draw on pencil, you have the structural shape, then you add the details.

 

I think you got subdivision modeling and sculpting mixed. Nothing prevents subdivision modeling to have good topology from the start. You just put subdivision modifier on top of lowpoly as a shortcut to make smoother shape. When the modifier is applied, the resulting topology will follow the structure of the base, just more dense. You can even use base lowpoly as one of the LODs. While sculpting is often used to make designs without bothering with good topology, often using generator tools like Dynamesh/Dyntopo as base; thats why in the end retopology is nessessary.

 

I think this is the best way to think about it. What does these different tools/methods actually do? When you understand that, then you can apply them however you need. You won't be limited to conventional wisdom.

And here's a thing about conventional wisdom: It's not always right! We spent a few days in my game trying to develop a conventional wisdom solution for a technical issue. Turns out the dumbest, simplest method that is basically the opposite of what conventional wisdom says to do was the best way.

 

For high poly hard surface models to bake from it's basically a necessity. Unless you are using weighted normals (which not all 3d apps have). even 10 bevels will not be enough to perfectly preserve the flatness of two surfaces joined by a sharp edge.

It is especially useful for hard surface highpoly models with a somewhat organic shape, such as the front of a car or a scifi helmet. You don't have to create a smooth curve, you just create a blocky shape and apply the subdivision.