Valfaroth said:
From what I have read, if you build it in pieces and put it together in the editor you are better off
if you have a column that is 20 polys....and you duplicated it 100 times....it only holds one in memory....if you build 100 columns at 20 polys a piece and make them one object....then it keeps 2000 polys in memory.
Don't know which way you built it but I thought i would throw in what i read....
While that is true the story is much more complex than just that. Someone picked up that info a few years ago when the tech was released but never gave people the full explanation.
Adding 20 seperate pillars will in fact have a perfomance penalty of at least 200-300 percent in most cases compared to having just 1 mesh that contains 20 pillars.
Theres more to a pc than memory. Splitting it in a single pillar and re using it will save memory (altho little since one mesh only takes up a few dozen KB at max most of the times) but it will eat cpu because the cpu and gpu will need to draw more objects.
Thats worse for your fps than what you gain out of optimizing stuff for the memory.
Every object you place requires the engine to do an extra draw call. Every material on a mesh is a draw call. Every material on every pillar is a draw call. 1 pillar with 3 different textures will be splitted in to three sections before being renderend.
Three times the same pillars means 9 sections to renders which is considerable slower than attaching the three pillars in 1 object. That would result in just 3 sections to render but a higher polycount as it wont be able to cull out the other pillars you dont see anymore. And so on.
Its a lot of balancing out. Theres no definite answer or quick solution as it depends strongly on the situation.
The amount of sections it has to draw is worse than the polycount fps wise. I can make 25 000 poly run like crap while a 500 000 poly building might run real fast.
I think Im going to write a tutorial on that somewhere in the summer. So many people overlook that or are misinformed.