[naikrovek]

it's not even a distance thing. that would be LODs which are apparently not done for human models.

Occlusion culling and frustrum culling are what are used to choose what is outside of the camera's view, because of either obstruction (a model is completely behind another and is therefore not visible) or is simply outside of the field of view of the camera. That turns whole models on or off, though, and not parts of a model.

The teeth are rendered, but because of how draw calls work, the performance impact of the teeth is miniscule compared against the overall performance profile of the entire game.

[ncallaway]

> The teeth are rendered, but because of how draw calls work, the performance impact of the teeth is miniscule compared against the overall performance profile of the entire game.

I agree as to the individual teeth, but in aggregate rendering the "high quality" model (thousands of triangles) for lots of small characters that usually appear off in the distance instead of the "low quality" model (tens of triangles) could drive a pretty non-trivial performance hit. As far as the teeth go, they're a pretty small factor overall, but I would imagine having no LOD on characters that are usually small and distant is a significant performance issue.

[naikrovek]

yes it is. I didn't say otherwise.

geometry instancing makes even complex models transfer to the GPU quickly (when measured per instance) when you send them all in the same draw call, but sending a few LODs in a few draw calls will take less time to rasterize, yes.

problems in this space are very fun to work on and think about, to me. that's why I work on completely different things. I'd hate it if I did it full-time.

[ncallaway]

Sure, I guess I was mostly responding to "the performance impact of the teeth is miniscule compared against..."

I might have just over-interpreted how small you meant by "minuscule". I'm not sure what other performance problems exist in that game specifically, too, which could impact it.