Interestingly, we actually can use normalmap compression schemas on this engine to reduce the compression artifacts greatly.
But it's never ever worth it vs. good ol' BC1 (DXT1), except maybe if you were doing something insane, like a million-tri thing you baked down to a smallish statuette or something just plain nuts like that The reason? Size in memory, unfortunately; it takes twice as much RAM to store BC5 NM vs. BC1 color encodings.
It's a real pity the engine loads textures into its main memory like it does and hasn't ever gotten compiled for 64-bit so that RAM isn't a huge limiting factor; it's kinda fun, though, using the engine to learn more about optimizing assets properly
For a sword like this, the handle's usually the last thing people look at, fortunately.
But it's never ever worth it vs. good ol' BC1 (DXT1), except maybe if you were doing something insane, like a million-tri thing you baked down to a smallish statuette or something just plain nuts like that The reason? Size in memory, unfortunately; it takes twice as much RAM to store BC5 NM vs. BC1 color encodings.
It's a real pity the engine loads textures into its main memory like it does and hasn't ever gotten compiled for 64-bit so that RAM isn't a huge limiting factor; it's kinda fun, though, using the engine to learn more about optimizing assets properly
For a sword like this, the handle's usually the last thing people look at, fortunately.