It is always better to pack these into the channels of a single texture, because you reduce the amount of needed texture samplers in the material, and because its much less video memory. Imagine you have a separated roughness, metallic and occlusion map. Texture still uses 3 channels but it has the same information in all channels. Or even if the other channels are filled black or something, they don't contain useful information but you load them for nothing. So they take 3x more memory than you actually need. By packing textures into channels, you avoid this, and you load only useful informations.
Replies