SSAA vs MSAA question

Roger Kohli

Newcomer
AFAIK, Multi Sampling AA is generally considered to be more efficient than Super Sampling (though having poorer IQ). Is this still the case if we are using 8 texture layers?
 
Roger Kohli said:
AFAIK, Multi Sampling AA is generally considered to be more efficient than Super Sampling (though having poorer IQ). Is this still the case if we are using 8 texture layers?

MSAA uses the same texture/color sample for each subsample. SSAA uses (potentially) different texture/color samples for each subsample. The savings goes up if you increase the number of textures in use (i.e. if you are using 4x AA and 1 texture, then you save up to 3 extra samples per pixel, in the case of 4x AA and 8 textures, you save up to 24 extra samples per pixel).
 
Dave B(TotalVR) said:
Yes, but both methods will be hit by the same % increase in fillrate requirements.

That's true, but his question was referring to increasing the number of active textures. Since the fillrate requirements are the same regardless of how many textures are active, then increasing the number of active textures make MSAA much more efficient. Of course, SSAA still gives better image quality.
 
Does this mean also that you only require one blend operation per additional texture layer per pixel with MSAA against four for SSAA (4X)?
 
Roger Kohli said:
Does this mean also that you only require one blend operation per additional texture layer per pixel with MSAA against four for SSAA (4X)?

It depends... If you are doing destination blending, then you will need to blend each sample in either case. However, if you are doing texture blending, you may be able to save computations in the MSAA case if you are able to handle multitexturing (because the texture blend is only done once per pixel, not per subsample). If you can't handle multitexturing at all, i.e. you have to write intermediate steps to the framebuffer, then you are taking a similar hit for both MSAA and SSAA.
 
Back
Top