fastest method for luminance texture

[zed]

im not caring about quality issues just speed.
A/ 8bit alpha/intensity texture (best quality results but perhaps slower than B,C,D)
B/ dxt5 (same memory as #A but i assume a slight overhead for unpacking, is this correct?)
C/ dxt1 (ie rgb as intensity)
D/ others?

zed

[zeoverlord]

A) should be fastest, the others only use less memory, but B and C do produce more texture reads than A, so they may be slower.
I don't know of anything that would fit into D, but the speed diff would not be that large compared to a corectly filtered luminance texture.

[Humus]

C) is fastest.

[plasmonster]

D) is slowest.

[tamlin]

Humus, just so I understand this right - using RGB DXT1 for intensity is faster than using a plain (should be 1Bpp internally AFAIK) intensity texture?

Is that ATI-specific (not included in OP's question), or is this performance statement true for all existing hardware and software implementations supporting DXT1?

[knackered]

I'd always assumed texture cache coherency is the reason compressed textures are faster that uncompressed. You can fit more texture area in the cache if it's compressed.