Let say I wanted to make a level-based game taking place inside a bunker. Would I just make the underlying level geometry as one 3d model including all walls, ceilings and floors, or would I divide it into models for each room, or even for each surface?
The most common practice nowadays is to build it with modular assets, and part of the games build process is to combine static meshes that share a material so it can be rendered more efficiently. This is a oversimplification of a complex topic, and there are far more advanced tools now. Most terrains aren’t even modelled anymore, but generated and sculpted in engine using PCG tooling.
Note, this is mostly for the contemporary AA and AAA games industry.
Depends on the game engine, and the game. But it’s usually not one big thing. Also keep in mind a lot of how a game engine imports the data could be doing quite a bit behind the scene. So, refer to the engine docs or communities.
From what I remember ftom modding and just playing games from a while ago, often there is “level geometry” (which can be subdivided into parts if needed), for example, ground and hills in Fallout, city block in GTA; and then there is “decor”, repeated assets, such as trees, fences, pipes, obstacles, etc. This asset repetition is particularly noticeable in Fallout 3/NV, and somewhat in Stalker (mostly on abandoned cars)
Usually modular.