I’m working out my level streaming setup, and in a lot of cases it would be made easier if we could afford to be somewhat aggressive with streaming distances and load nearby areas when you get anywhere close on the off chance that you turn and move in that direction. However, I’m concerned about how much that’s going to inflate the rendering pipeline. Right now we’re only holding and drawing a maximum of two levels in memory: the level the player is occupying, and the level they’re looking at/transitioning into. It would make level design massively easier if we could keep the 4-5 closest levels pre-emptively loaded into memory, the idea being that 90-100% of each individual level’s meshes would be concealed by frustum or occlusion culling, but I don’t want to gouge the game’s performance simply to make level design easier. That having been said, how much does a mesh that frustum or occlusion culling has hidden actually take from the system?