There are many different ways of sampling and all cause different result to the final image. To prove how big off a difference the sampling part of anti aliasing makes some comparisons between random sampling and grid sampling was created as can be seen in the images bellow. Random sampling causes the image to look fuzzy, while the grid sampling makes it look more crisp. The fuzzyness of the random sampling is a side effect of not sampling diversely enough to get a good representative color of that specific pixel.
Comparing random sampling with grid sampling proved that sampling patterns really do matter for the end result of the rendered scene. But how does two sampling patterns that are not random compare to each other? Bellow is a comparison between the Grid sampling pattern and the Quincunx pattern. The Quincux sampling creates smoother edges while the grid sampling as stated above creates crisp edges. Note that this was not Nvidias Quincunx method which shares neighboring pixels for a performance boost that was implemented, but just a normal quincunx pattern for each pixel.[2]
 |
| A Quincunx pattern |
There is no perfect sampling pattern though there is some that are suited better than others for certain scenes. Increasing the number of rays sent out helps creating a more diverse sampling pattern and creates a more representative color for that certain pixel. Increasing the number of rays increases the rendering time substantially.
References:
[2]HRAA: High-Resolution Antialiasing Through Multisampling, Technical
brief, NVIDIA Corp.
http://www.nvidia.com/attach/151 (Accessed 2015-05-29)
http://www.nvidia.com/attach/151 (Accessed 2015-05-29)
