science of texture size

[Rhiannon Arkin]

Hi

I was just wondering, artistic choices aside, is it load weight wise better to make 1x 1024 texture to cover lets say various parts of a building or to have 16x 256 textures? 

or is it just the same ? 

 

R.

 

 

[ChinRey]

The difference is marginal but in theory at least a single 1024 should be slightly lighter than 16 256's. I addition to the actual pixel colors, each file also includes some headers and other auxiliary data and a single big file is also usually easier to compress than several smaller ones.

[Drongle McMahon]

In addition to the per-file overhead, the multiple texture option also involves more transactions with the database and server. It will also allow finer distribution of avalable texture area between faces according to their sizes (unless you need only fixed size sub-textures.

However, there are a couple of considerations that might swing the comparison the other way. You should avoid mixing alpha and non-alpha areas in the same big texture, as doing so will use more data for the latter (32 bit/pixel vs24) and will probably cause alpha sorting artefacts with the textures that don't need alpha channels.

Otherwise, it may depend on how your use of the textures is likely to be distributed among multiple objects inworld. You could be downloading multiple redundant texture components in the 1024 if they aren't always going to be used all together. There might also be complications if you want to allow users to change textures (either for you or for the user).

Finally, your composite 1024 texture won't be tileable, and tiling small textures can be one way of improving efficiency of resource consumption. Again, that depends on the exact application.

 

[Rhiannon Arkin]

I would have guessed that one texture might be less data handling then many small ones. 

But you have good points about tiles. they are quite useful at times. 

I have not yet had issues with mixing transparent and non-alpha parts in one texture, using png and alpha blending. 

I turn it off for the non-transparent material and leave alpha blending on for the transparent one. Seems to work ok. And i don't think that png is never not having some sort of alpha in it. i could be wrong. 

 

[Drongle McMahon]

True - you can avoid the alpha problem with the (new) alpha mode settings. But - False - png format can be 24 or 32 bit (i.e. with or without alpha channel. Not sure how you tell photoshop, but in Gimp I just merge image to one layer and remove the alpha channel. Then it automatically exports png as 24 bit without an alpha channel. You can see that both in the export dialog and after import where there's no alpha mode options.

[Rolig Loon]

In Photoshop, the Save For Web option with a PNG file saves it as a 24 bit image.  A standard Save creates a 32-bit image.

[Rhiannon Arkin]

oh great. it saved me 160kb. 

 

[Arduenn Schwartzman]

My guess is that 16 x 256x256 loads faster (because it happens in parallel) than 1 x 1024x1024. But after that, it'll render slighty more slowly, because he viewer has to handle 16 source files rather than 1, while rendering a frame. So the question boils down to: do you want it to load faster, or to run smoother?

[Cube Republic]

If you wish to get super efficient, and some elements of your builds are symmetrical, you can mirror 2 or 4 ways. Think of a table top, an arch or other symmetrical objects. Often this gives a crisper result because you can effectively double the texel density without adding more textures. It's good to add a bleed edge around your islands if this method is used to avoid seams as the texture  mip-maps. 

Edited March 15, 2017 by Cube Republic