I wanted to add some stuff here, but before I started editing mercilessly I wanted to get some feedback…
I’m pretty sure DXT is compressed in 4x4 blocks, not 2x2.
“All DXT algorithms are supposed to compress textures with the same algorithm for the color component”… not exactly true, i.e. DXT1 is different bit depth per pixel than 2/3/4/5.
I based about 10% on the Mathis stuff, because yes, it is anecdotal and inaccurate, realized that the more I looked into it.
4x4: You are right, my bad.
Color component: Every article I’ve read on DXT color compression has said all DXT RGB channels compress the same. In practice, especially with the NVidia PS filter, I’ve noticed this isn’t the case, but I’d like some more concrete info.
I have sort of wiki-fatigued myself, but will try to add some more stuff this weekend and look the DXT article over if you don’t have time. We’re also busy on getting the script versioning up so I may not have time, we’ll see.
DXT compression uses an algorithm called: EPht Compression and it works like this:
I think the image is pretty straight forward in explaining exactly how this compression works, but in case its not clear from the image, it involves an elephant and some peanuts.