*Pots* wrote:
Psychic_313 wrote:Also, please don't make any textures over 256x256 pixels. The Unreal engine used in UT has a
hard-coded limit of 256 pixels, and will resize anything larger to fit. If you use larger textures,
you're just wasting memory and bandwidth for no benefit - you might as well resize it yourself
before you import it.
I only say this because I've seen one skin including a 1024x1024 texture. Hmm, 16 times the normal
file and memory size - ouch!
Now not that I care that much, but bigger textures = bigger file size = waste of time and space. So who's right and who's wrong?
Wrong. I don't know about the old Direct3D rendering, but at least the ones that came after don't have that problem at all. Either you use them in BSP or meshes, you get indeed more detail when you import larger textures, although I personally would advice the limit of 512x512 textures for meshes (limit I personally use for my weapons and vehicles in UT, and they get indeed much more detailed than if I used plain 256x256 ones) and a limit of 1024x1024 textures for mapping (I used 2 or 3 for an Egypt themed CTF map I made last year).
Plus I underlined the "bandwidth" because, clearlly, that guy doesn't know how this thing works and was been told by some rumour how it's supposed to work. So basically, even if you had a colossal 10.240.000x10.240.000 texture, it wouldn't affect on anything in bandwidth. Why? Because the texture is not replicated, it's downloaded instead. It only rises the filesize a bit, but even so, downloading a 100kb or a 1Gb file from a server doesn't affect "bandwidth", it just accupies some space of it during more or less time, and it keeps somewhat the same, so the "bandwidth" usage amount is not affected by filesizes nor texture resolutions. So textures don't affect bandwidth anyhow.
Regarding memory, yes, the bigger the texture, more memory it will occupy, but in that case look at our machines nowadays, any decent VCard can save all the UT textures with no space or speed problems, so you don't have to worry with "memory".
All you have to worry about is filesize. The bigger the filesize, more time it will take to load up from the HDD or download it from a server. So when mapping, if you're using a custom 256x256 texture, and you will use it always scaled down in the whole map, resize it to 128x128 or lower instead (depending on the scaling), and it will save you space on your file.
The same with skins and HUD elements: if you don't need it to be so big, resize it or make it small when doing it from scratch. For instance I have 512x512 both in weapons and vehicles, but I only use 512x512 skins when I really must o, otherwise I make my UVs so I can create the skins as 512x256, or 256x256, or even 128x128 or even lower than that, depending on how much detail I really need. Otherwise my filesizes would jump from 80Mb to 200Mb, and take a sh*t load of more time to load up.
UT doesn't have anything like that hardcoded afaik, what can limit the texture resolution is the renderer used, but generally the old D3D crashes instead of resizing, unless it has mipmaps.
Also, you can save memory and processing enabling mipmaps on texture import. Well, in case of 256x256 textures and below for SKINS and HUD, it should have indeed mipmaps disabled and lodset=2 and lodset=0 respectivelly (skin detail or no detail algorithm option, and belive me or not, it makes a use difference in texture detail on a mesh or HUD), but anything above 256x256 or even for mapping has to have ALWAYS mipmaps enabled, no matter what's used for.
In case you don't know what mipmaps are, they are "texture clones" with lower resolution from the original texture, so let's say you have a 256x256 texture and mipmaps enabled, it will save also a 128x128 copy, 64x64, 32x32, 16x16, etc, within the texture itself, used mostly by BSP surface processing when you're far from a surface, so instead of showing up the whole 256x256 set, since you're looking at it far away it might be showing its 128x128 or 64x64 version, avoiding a jittered look (that can give some headaches), and still look detailed enough, somewhat a bilinear filtering.