Maps Question/Concern

DavidCruz

An odd one to ask for sure but was wondering as i thought i read on pc that we could use alternative sizes for the maps we use?

Like can i have a 2048 Normal map and have all other maps be 1024?

^Pretty much that can we get away with this in an actual game engine?
Of-course i can on marmoset and other things such as this but a functional practical process is more what I am after with asking this, I.e. will doing this across a vast majority of models make a game engine slower with a higher normal map?

Or would it just be easier to have all the maps be 2048? I am trying to make the most out of 1024 which is why i am asking this, but doubt a 1024 normal map will do my project justice.

Thanks

throttlekitty

Yes, you can mix and match sizes, compression too if you're using DDS.

EarthQuake

Yes, and further, sometimes it can make sense to use 1K uncompressed normal map rather than a 2K compressed normal map (if your engine supports uncompressed textures) because the VRAM use will be the same but sometimes compression artifacts hurt worse than the lower res texture.

DavidCruz

I didn't even know about uncompressed textures, so much to learn.
Is all this info on the wiki? I usually pass it up, regrettably.

Would love to test a few uncompressed 1k N Maps, will check the wiki hoping the answer is there.

Thanks gents.

edit:
Oh so compressed textures is just the blue channel removed and all textures are uncompressed? wiki info.

throttlekitty

DavidCruz wrote: »

I didn't even know about uncompressed textures, so much to learn.
Is all this info on the wiki? I usually pass it up, regrettably.

Would love to test a few uncompressed 1k N Maps, will check the wiki hoping the answer is there.

Thanks gents.

edit:
Oh so compressed textures is just the blue channel removed and all textures are uncompressed? wiki info.

Not quite, that's a packing method that skirts around some of the issues that arise with compressing a normal map. (this requires a shader to rebuild the blue channel at runtime, fyi)

You're probably familiar with jpg compression options: 0-100, lower settings net you blockier or smudgier images. With the .dds format you have different types of compression methods you can use, DXT1,DXT3,5,DXTN(normal),etc. Each of them more or less works the same way but with different bit ratios and whether or not an alpha channel should be present. 8:8:8:8 is an uncompressed dds type, but the filesize is larger. For normal maps that don't rely heavily on high frequency detail, we can usually get away with a smaller resolution image with no compression and avoid artifacts.

DavidCruz

Thanks all, I came back prepared instead of just asking questions without trials to show.
Tbh it's making me batty now.
I also did read this.

So is it just best to go with a 16bit and drop attempting to create a decent 8bit with no noise or banding or distortions as the examples?

I would just go with the 16bit but i want to learn as I had no idea i was creating 16 bits when most engines only support 8, or has this changed by now for most?

Thanks again for any help, this took a while..

Then i saw this and i was like how did he...
http://www.polycount.com/forum/showpost.php?p=2142854&postcount=10


Bipidabobpodi-boo, this is xnormal 2048 bake to resize 1024, to 8bits, normalized, i will accept these results, unless i find something else.

Edit:
I made it simple here:
https://www.artstation.com/artwork/8bits-tga-how-to-very-short-tut