How do you measure AA ?

[rms]

Many times I've seen mention of ATI's 4xAA being superior to Nvidia's 8xAA (or some similar claim). Is there any method of measuring if in fact this is the case, that (for instance) the Nvidia drivers are simply lying to the user about what level of AA is actually being displayed, or are using an inferior algorithm that doesn't qualify (to some accepted standard) as the reported level?

How do you measure AA? It seems to me that there is unlimited room for videocard manufacturers to slap something on the screen and call it AA or FSAA of whatever level, if there is no way to verify their claims?

rms
rsquires@flash.net

 

[BRiT]

Nothing quantitative, only subjective -- run a few games at the desired settings and compare image quality, as well as running the wonderful FSAA tester program.

The drivers are not lieing about the levels being used. The reason Nvidia's AA looks inferior to ATI's AA is because of the vast differences between ordered grid and sparse gride, combined with the lack of gamma-correction and presence of gamma-correction.

 

[Pete]

With my eyes. And if my eyes tell me that ATi's 4xAA is superior to nV's 8xAA while being much faster, then my brain concludes that nV's (hardware) implementation must be inferior.

But I'm just speaking from screenshots. I can't say from personal experience how they compare in motion, but I'm not reading good things about nV in that regard, either ("blurry").

 

[gkar1]

Rotated grid multisampling(ATI) is superior to ordered grid multisampling(NVIDIA) in almost all cases. Also the addition of gamma correction in ATI's AA is another point in favor of their implementation.
Its not only a matter of perception.

 

[Simon F]

How do you measure AA? It seems to me that there is unlimited room for videocard manufacturers to slap something on the screen and call it AA or FSAA of whatever level, if there is no way to verify their claims?

rms

RMS,
pardon the pun, but you could use the RMS (root mean square) of the error (i.e. difference) between a very highly supersampled and correctly downfiltered image and that of the technique of graphics card being tested.