One saves time and energy (and spends less triangles), the other saves uv space/texture resolution. Comes down to what the asset is used for, how much screenspace would it take (size of asset), and maybe how important and closely viewed it would be.
Thank you for your answer, but I still can't quite figure it out, could you help me with this? Write more detailed or maybe you could give the material for reading or familiarization, so that I could understand, since I did not find the information in Google.
Why something is modeled one way in one example and differently in another often comes down to the specific technical limitations or requirements of each project. Both approaches are used to optimize a specific aspect of the model. Eric's summary is a pretty good explanation of the basic concepts of why an artist would tend to choose one approach over the other.
One drawback of overly broad technical questions is they often lack the context required to provide specific answers that cover every situation. More often then not the answer to these types of questions ends up being "It depends."
A good place to start would be looking at the Polycount Wiki entries for texture baking and polygon counts. These entries link back to additional documentation and related topics that cover different optimization strategies and reasons why certain technical decisions make sense in specific applications.
From there it would make sense to search for terms like: contiguous low poly mesh, overlapping geometry, intersecting mesh objects, low poly optimization, etc. There's a lot of topics here that cover these topics and explain the how, when and why of it all. Here's a few threads that are worth taking a look at.
The links above are listed in chronological order so it's worth noting how some answers change over time as the technology evolves. In the past it may have made sense to ruthlessly optimize for lower
poly counts but with normal baking and advances in render engine
technology it often makes sense to take a more balanced approach
that focuses on improving overall visual quality.
Both contiguous and intersecting mesh components can be viable optimization techniques when used appropriately but it's also important to avoid over optimizing for one specific geometry element as this can cause unnecessary issues for baking and rendering.
@FrankPolygon WOW just WOW. I have no other words, you helped even more than I could have imagined, I am grateful to you for all the material provided. There is many new thing and very interesting material to learn. And thank YOU for the helping me and warm welcome!
What really grinds my gears is when I get to see a lowpoly teeth model where all the teeth are individually modeled chunks, stuck into a lowpoly gum mesh where you can clearly see the intersections everywhere. And of course unwrapped as unique islands.
The most likely 'technical' reason I can come with for that is.... some people are just sloppy. :P
I usually merge meshes together wherever I can but there are a few exceptions.
Reasons for merging vertexes:
Merged vertexes are much easier to bake maps from as you don't have overlapping geometry on the lowpoly. This is a big time saver and produces much cleaner bakes.
Merged vertexes have fewer visible hard edges. Hard edges are visible on a model as un-aliased edges and look bad - even with high end screen space anti-aliasing.
Merged vertexes save texture space as you can UV the mesh as a single continuous UV island - which is always less UV space and saves texture space. This can lower the vertex count of a model as well.
Merged vertexes produce better LODs if you are using a automated LOD generation system like in Unreal. Having separate shells means the mesh will 'tear apart' when dropping to lower polygon LODs.
Merged vertexes produce cleaner lighting as the polygons are more evenly sorted.
Reasons to separate shells:
Separate shells are slightly faster to model and so as long as you're not baking maps you save a tiny amount of time.
Separate shells do save polygons and in some cases, like modeling a fence or grille, can save a lot of polygons. But they don't save vertexes so it's more important for low minimum spec work like phone titles.
Separate shells when you want pieces to move independently of each other in some way - like for a gun or animated prop.
Separate shells can make it easier to make manual LODs as you can just delete shells to lower the polygon count.
Replies
http://wiki.polycount.com/wiki/Polygon_Count
Reasons for merging vertexes:
Reasons to separate shells: