Precisely. Hardware that does not gamma correct the sampling is actually skipping a step that is necessary to correct the blending for the non-linear response of the monitor - blending the samples in linear space does not produce a linear colour transition when viewed on screen.DaveBaumann said:Any image that is displayed via a CRT has a gamma ramp applied to it - just averaging FSAA subsample values outside of gamma space will give an image that is less appropriate for the display. Gamma corrected AA is actually giving an image closer to what is supposed to be displayed and thats why people view it as looking "better" as it is actually more correct than AA without it! You should be questioning systems without gamma corrected AA as these are "altering the image" (in an incorrect way) more than those that have the gamma ramp applied in the subsample weighting.
As far as I know there is no strict specification in D3D for how the samples used for multisampling should be combined by the hardware. The only obvious note about it in the documentation is:
"The various samples recorded for each pixel are blended together and output to the screen"
As such a linear or gamma'd blend of the subsamples can be viewed as 'correct' from the API point of view, but the gamma corrected version will look better.