system:
intel i7 920
geforce 275 gtx
6 gb ram
---

Well, thing is, in veiwport, i have like 100k triangles, so it runs smooth as it can get, though i have like 1/2 of model hidden (so i'd say total scene size is like 250k triangles).

What i am doing is, to make high poly model (what of those pillars have 200k+ polys), and then low poly, that still has like 10k polys. I then use mental ray proxy, so that in viewport, you see this low poly version, to avoid unnecesarry memory usage, and when i render it, mental ray replaced low poly with high poly. this way, viewport runs quit smooth, and render times are a bit faster.

So, in short, in viewport, i actually have 200k tris or less (depends how much that low poly has - it can be just cube as well, but i prefer to have model that is very similar to high poly), however, MR actually does render 2 milion polys (or so).

Also, i made only one MR proxy pillar, and then duplicate instance. This way, atlest from what i see, is best to achiv faster rendering time and reduce memory usage (together with proxy), while viewport runs quite smooth.


tnx for comment!