So for performance reasons (and maybe also because it theoretically could be a disk space issue), can I get a quick-and-dirty answer to this kind of a scenario:
I'm working on a environment piece currently that's a witch's house. On the top of the house, there will be moderately detailed wrought iron weathervane (y'know, kinda like the ones that have the chicken on them and then the arrows etc.)...if I want this weathervane to look really nice, I could use a relatively large (for the object at least) set of texture maps with a high-resolution alpha and just use planes with some displacement on them for thickness. The geo and vert count would be really super low but the texture maps would be fairly large to ensure great rez. I'm thinking 512x which doesn't sound like much maybe but it would be just the lousy weathervane and nothing else which might be excessive even for a house that features a set of 4k maps for EVERYTHING else about the house.
Alternately, I could just model more of it in which case the texture maps could be really small but now my geo is more complex and has a much higher vert count.
Which is better these days? Which way would you all go? Bigger maps or more geo?
Replies
POM would likely be a better option than displacement.
Higher vertex count is generally better than more expensive shaders. Most games are fill-rate bound these days, so reducing the amount of draw calls and reducing alpha blend/alpha test is generally the way to go.
Conversely, optimize one bad material and you can get back far more frame time.
You will rarely ever be told to get rid of tris/verts to make the game run faster, for the most part that is a fools errand. Art optimization mostly comes from draw calls/shader optimization/screen space fx optimization.
That doesn't mean go haywire with your verts/tris, but if you are targeting modern advanced hardware, be liberal with verts/tris. If it isn't contributing to the model, remove it!
big textures though, that kills me especially if you are streaming.