To Stitch or not to Stitch?

[Free_Login]

Hi
Please help me decide - to stitch or not to stitch. (We are planning to use this model in Unreal 3 game engine.)
You can see the situation in the pictures. There is a wall and an arch. In the first case the arch is stitched into a wall, which leads to the formation of additional triangles. In the second case, the arch and the wall are two different meshes combined into a single object, thus reducing the number of triangles.
There are conflicting opinions at the moment on what is best in this case. Any suggestions?
Arch and Wall

[Nawrot]

Not sure if you seen this tutorial, but i think it will explain some aspects: http://www.hourences.com/tutorials-u...ng-2-advanced/

And as to your question (if you cannot make up your mind after above tut),

all really depends on what you are making and where you are aiming at.
For best details, i would make each column separate. Make model of 1 column+2 halfs of arch on each side, then repeat it, add some material variation on top of it and vertex paint to hide fact it is same model copied. This allows biggest poly count and best texture details (1024x1024 is used for one arch not whole section).

medium solution is your unstiched idea, you can reuse groups of arches more than whole wall+arches.

And stiched imo is making stuff like in max for rendering, you are not using any of engine tricks to raise polycount and texture density.

Also there is one major reason for splitting all: dynamic lights (it is explained in tutorial), if you want use more than one dynamic light and its possible that more than one will lit that huge mesh, better split it for performance reasons.

[Sir. Polaris]

What Nawrot said is correct. Recycle as much as possible and keep things split, one for reuseablity and two so when the engine wants to load and drop things and render things on things, your doing so with the smallest possible parts.Increasing efficiency and making it possible for you to crank up the jam on the poly count and texture sizes.

[Tom Shannon]

They are both right... to a limit. Modularity and reuse-ability are great until you start getting into very big environments and big draw call counts, but I think that's a discussion for another day. You're asking about the wall and if you should use fewer polys and a less accurate model or more polys and a more accurate model.
I generally prefer modeling a 'stiched' model. The 'stitched' model will provide better lightmapping and fewer flickers in general. the increase in polys is generally not too much. However I've had architectural treatments that when modeled 'stitched' end up being 4-20x (!) the polygon count so we decided to go with intersecting geometry.