From http://opengl.org/registry/specs/EXT/texture_sRGB.txt

Quote:8-bit components of sRGB texels are converted to linear RGB values which requires more than 8 bits to avoid lose of precision. This implies the filtering involve more than 8 bits of color precision per component.

So it seems that every time the texture is used, the value is de-gammaed into a 16 bit format. I'm wondering what kind of tradeoff this extra processing is in comparison to manually de-gammaing non-HDR textures and storing them in 16 bit (since putting de-gammaed sRGB into 8 bit linear loses some bits in low brightness parts of the dynamic range). Is extra processing likely to be higher or lower cost than the extra memory of a 16-bit per channel texture format?

Hello?

It's impossible to say, as it is completely implementation dependent and can vary widely from one 3D chipset to the next.

Just try both and profile to know how performance behaviour is on your GPU. Just keep in mind that it might be entirely the opposite on some other GPU. That said, it's very likely that you won't even notice any difference between both, unless you are seriously fragment limited. And even then, it's likely that the texture transfer bandwidth and cache behaviour is going to have a much higher impact on performance than some cheap gamma correction.

Thanks.