UVs: More Maps VS. Bigger Maps?
polycounter lvl 10
Say you're working on a character model with several components to it, that you are trying to UV unwrap efficiently. In terms of maintaining good texture quality (and good performance), would you rather:
1) Stick in all on one map at a very high resolution (e.g. 4k+)
2) Divide it between a couple of medium resolution maps (e.g. 2048x2048)
3) Divide it even further between 3 or more maps at a lower resolution (e.g 1024x1024).
I guess my question boils down to this: which has the bigger performance hit, more maps or bigger maps?
1) Stick in all on one map at a very high resolution (e.g. 4k+)
2) Divide it between a couple of medium resolution maps (e.g. 2048x2048)
3) Divide it even further between 3 or more maps at a lower resolution (e.g 1024x1024).
I guess my question boils down to this: which has the bigger performance hit, more maps or bigger maps?
Replies
However, in some cases, like skin vs cloth, you might need multiple shaders any way, at which point, having a map for all the skin elements, and another for the non-skin elements won't make a difference vs 1 map for both. If your skin shader requires an extra map like a subdermis or translucency input, its more efficient to have all the skin content on a smaller texture separate from everything else, so you don't end up with texture that have large black wasted space for areas that do not require those effects. Sometimes these secondary effect maps can be significantly lower resolution than the base diffuse and normal maps.
Similarly, if some elements of your texture need an alpha channel for transparncy, you should generally group all of those elements onto a separate, smaller texture. This is important for two reasons, the first of which being you don't want to run alpha blending on more polygons than you need to, secondly is texture memory/compression(see below).
In terms of texture memory, 1x 4K is equal to 4x 2ks, 16x 1ks or 64x 512s.
Texture compression and alpha channels are also very important. It might seem really efficient to pack your specular map into your diffuse alpha, but in reality it isnt. Compressed, a 32 bit image (RGB+A) is the same size as two full color 24 bit images.
Now, if you want to get really efficient, what you should be doing is packing multiple textures into the channels of a 24bit image. For instance:
R: gloss/rougness
G: specular/metalness
B: ao/cavity
TL;DR: It depends on exactly what you are doing!
How the hell do you do that?!
In Toolbag, you can load the same texture input into multiple slots and then use the channel selector to pick the right one.
In unreal, you can link individual channels to inputs in the node editor.
I think UDIM style texturing will be a good future direction once someone writes a good texture array implementation. Film assets are often using just 512x512 textures, just very very many of them.
Basically there's no workaround to get this in games that don't support this? I'm a modder and haven't really gotten into the "big thing". I was really excited when I learned about this.
We are using UDIMs in Mari too for our tv spot productions and 4k textures are our standard. For some characters or objects we use 8k UDIM tiles. It has no impact on the rendertime because arnold (and some other renderers) use mipmapped textures.
Yeah, that's a very good point. The larger the texture is the harder it can be stream in and out of memory.
I've gotten some film assets set up like this, though they might have been scaled down for other reasons.
I've noticed that projection speeds can be kind of slow with 4K tiles in Mari if you use heavy meshes, I'd rather go 2k or lower and have a different unwrap. But I'm not super experienced with this kind of workflow.
From an artist perspective it might be a good idea to work with more smaller texture, from an engine perspective this could result in more rendering calls, which is still an issue (Mantle,Dx12 might change this in the future).
On the other hand, some game projects, regardless of the underlying engine, utilize (project specific) tools which will merge smaller texture into larger one (atlas). It really depends on the project. Without any project specification it would be a good idea to keep the number of textures down without sacrifying too much workflow or using uber-large textures.
i'd also think that character's materials take comparably small resource footprints next to environments or effects in a typical game. unless your place has hired the kind of noobs that put 2k's for eye textures and zbrush everything up to 11. not really something to split hair about early on. getting them to look right seems to be the main issue.