hi, looking at the paper by id/nvidia on compression
http://developer.nvidia.com/object/r...mpression.html

I wondered which kind of compressions are used in the real world.

3DC/DXN/LATC/RGTC would be same

I assume you consider regenerating the blue channel spare channels are always usefull (spec power, mask etc). We use dxt5 with xy in ga.

I voted DXT5, but we actually use the DXT5nm option in the nVidia DDS exporter.

LATC... so, 3Dc I guess.
I think we still use DXT5 on some systems but are phasing that out where possible.

DXn on 360.

thanks guys, and yes I assumed the 2 component trickery for tangentspace and DXT5, as the id paper also showed the results are best. DXn/LATC would be same/similar as 3Dc.

Compression is so last gen.

I was trying to convince of 3dc, but failed. You know how programmers are if they think "its good enough". That, and we have the specs alpha in the normal maps (long story and bleh engine), which Im not sure 3dc would even allow as it is.

Results vary. Look fugly to me, but ehh.

3DC and alike are a two component format only. Which means X & Y of the normalmap is stored, and Z can be reconstructed in tangentspace, as it will always "face up a little".

Depends on the kind of texture, but DX5 in general.