Anti-aliasing technologies

Definitions

	Aliasing Aliasing is any artifact that appears when lowering the resolution of an image. These artifacts typically appear as jagged and flickering lines and edges.
	Temporal aliasing Temporal aliasing refers specifically to the flickering or shimmering effects that aliasing causes when objects move in a scene.
AA	Anti-aliasing Anti-aliasing is any technology that is designed to minimize aliasing and temporal aliasing artifcats. Anti-aliasing is relevant in a 3D graphics rendering pipeline because part of the pipeline handles graphics as geometry of essentially infinite resolution. In a late stage in the pipeline, known as ROP, Raster Operations, the color of each pixel on the screen is determined based on the geometry. It is in or after this stage that any anti-aliasing is performed.
	Sampling Sampling is a term borrowed from signal processing. Sampling is the process of selecting a single value from a continuously varying series of values. The result of a sampling is a single value called a sample. In some instances, the term is used somewhat loosely. In 3D graphics, a sample describes a point-sized (infinitely small) area of geometry. When rendering 3D graphics, the challenge is to use the point sized samples to color pixels, which have a physical area. The more samples are available from the various resources that make up a 3D scene, the more accurately the pixels can be colored.

Anti-aliasing based on multi-sampling

Overview of AA methods that work by calculating extra samples and then using those to more accurately select the color of each pixel.

FSAA	Full Scene Anti-aliasing Describes any AA technology that works on the whole screen. This is as opposed techniques that work on parts of the screen, such as TrAA.
SSAA	Super-sampling Anti-aliasing This is the type of AA that creates the best image quality. It is “pure” AA with no compromises. All the other AA technologies exist because SSAA is very slow. SSAA works by calculating multiple samples for each actual pixel on the screen and then averaging those samples to determine the color for each pixel. The ratio of samples to pixels is often described by a number put before the acronym. For instance, 8xSSAA describes that each pixel on screen will be the avarage of 8 samples. When there is no multi-sampling, the color of each pixel is determined based on a single sample from the center of the pixel. When multiple samples are calculated per pixel, those samples are distributed across the pixel in patterns designed to maximize the AA effect. In particular, this means that the samples are not evenly distributed and do not line up horizontally or vertically. SSAA is slow because the pixel rendering stage of the rendering pipeline must do the same amount of work as if the screen resolution was many times the resolution of what it is. For instance, with a resolution of 1920x1080 and 8xSSAA, the hardware must calculate the same amount of pixels as if the resolution was 7680x2160.
OGSSAA	Ordered Grid Super-sampling Anti-aliasing (OGSS) Same as SSAA.
RGSSAA	Regular Grid Super-sampling Anti-aliasing (RGSS) Same as SSAA. Note: RGSSAA has two separate meanings. See the entry below.
RGSSAA	Rotated Grid Super-sampling Anti-aliasing Similar to SSAA. The difference is that instead of the sample grid being along horizontal and vertical axis, the grid is rotated to a fixed angle. This causes better anti-aliasing of edges that are almost horizontal or vertical. Note: RGSSAA has two separate meanings. See the entry above.
SGSSAA	Sparse Grid Super-sampling Anti-aliasing Similar to SSAA. This type of AA is the first in a series of AA types that represent a compromise designed to trade image quality for performance. The sampling pattern of plain SSAA is called Regular Grid or Ordered Grid. It can be pictured as a grid that has higher resolution than the pixel grid. Where SSAA samples in each square in that grid, SGSSAA samples in only some of the squares.
JGSSAA	Jittered Grid Super-sampling Anti-aliasing Similar to SSAA. The difference is that sample locations are altered so that they no longer exactly follow the grid that plain SSAA uses. This helps hide regularly occuring artifacts.
	Stochastic Super-sampling Anti-aliasing Similar to JGSSAA. The difference is that the sample positions are fully random instead of being just randomly offset from the grid as in JGSSAA. The white noise that this introduces helps hide rendering artifacts.
MSAA	Multi-sampling Anti-aliasing In a pixel pipeline, the final color of each pixel on the screen is determined by taking into account samples from Shader, Texture, Color, Depth and Stencil resources. Of these, Shader samples are by far the most expensive (slow to calculate) because the source of these samples are calculations done in software by an arbitrarily complex program that runs on the graphics card, called a Shader. Texture samples are also somewhat expensive. So, while SSAA retrieves the same amount of samples from each resource, MSAA lessens the load on Shader and Texture resources by only retrieving a single Shader, Texture and Color sample for each pixel regardless of how many Depth and Stencil samples are retrieved for each pixel. The result is that, compared to SSAA, MSAA runs much faster but it smoothes only the edges of polygons. It does not help for the internal areas of polygons as those are based on samples from Shaders and Textures.
CSAA	Coverage Sampling Anti-aliasing Extends MSAA. It adds a type of samples called Coverage samples. The grid of Coverage samples is of higher resolution than the grid of samples used by MSAA and provides a high resolution yes/no indication as to if a part of a given pixel is covered by a polygon. By using this high resolution grid of samples together with the other samples already handled by MSAA, CSAA is able to more accurately determine how to color each individual pixel. Coverage sampling is less expensive than it may first seem because the yes/no indication can be encoded in a single bit. This means that Coverage samples use much less memory and memory bandwith than MSAA samples.
TrAA	Transparency Anti-aliasing Describes AA types such as TMAA and TSAA that are designed to handle flat pictures with see-through areas (Billboards, Decals or Point Sprites in 3D graphics lingo) differently from real 3D models. In a 3D scene, there are two main ways of displaying items that have see-through areas, such as a tree. If the tree can be seen from all angles, a 3D model is used. But if the tree can be seen only from one angle or from far away, one may be able to get away with displaying just a flat picture of a tree in that location, thereby rendering the scene faster. TrAA applies to such flat pictures. TrAA exists because flat pictures are not handled well by FSAA methods other than SSAA. This is because flat pictures do not have any geometry associated with them and AA methods other than SSAA derive most of their samples from geometry.
TMAA	Transparency Multi-sampling Anti-aliasing A type of TrAA. It works by applying MSAA to flat pictures while leaving the real 3D models in the scene to be handled by other AA methods or by MSAA with a different number of samples.
TSAA	Transparency Super-sampling Anti-aliasing Like TMAA, but uses SSAA instead of MSAA for the flat pictures.
AdAA	Adaptive Anti-aliasing A combination of TMAA and TSAA.
HRAA	High-Resolution Anti-aliasing This method was introduced in GeForce3 in 2001 and was perhaps NVIDIA’s first stab at improving the performance of basic SSAA. In this method, the two sampling locations of 2xSSAA are adjusted so that one sample is in the upper left corner of the pixel and the other is in the center. Thus, when looking at a grid of pixels, each pixel ends up with a sample in the center and one in each corner, for a total of 5 samples. HRAA then uses all 5 samples to determine the color of the pixel. In other words, each corner sample is reused by all of the 4 pixels that share that corner. The technique gives quality similar to 4xSSAA with performance like 2xSSAA.

Anti-aliasing based on post-processing

Overview of AA methods that work by post-processing. As opposed to the multi-sampling AA methods, these methods work by adding a final step in which an algorithm examines the rendered image and adjusts pixels to reduce aliasing artifacts. In a pure post-processing based method, no extra samples are calculated. All post-processing based methods can be combined with multi-sampling methods to leverage the strengths of both.

Post-processing based AA has become possible due to the increase in computing capabilities in modern GPUs.

MLAA	Morphological Anti-aliasing The first AA method based on post processing. Invented by Intel and adopted by ATI for use on Radeon graphics cards. Algorithm: Find discontinuities between pixels in a given image. Identify predefined patterns. Blend colors in the neighborhood of these patterns.
FXAA	Fast Approximate Anti-aliasing NVIDIA’s answer to MLAA. The general consensus is that FXAA does not look as good as MLAA, but that it has less impact on performance (3-5%). FXAA does not directly address temporal aliasing. It may also not anti-alias all edges in a scene.
TXAA	Temporal Approximate Anti-aliasing TXAA is designed to minimize both regular aliasing and temporal aliasing. In motion, TXAA approaches the quality of AA algorithms used for CG graphics in the movie industry. Algorithm: NVIDIA has not disclosed the details of the TXAA algorithm. They describe it as a combination of hardware anti-aliasing (traditional multi-sampling methods), a custom anti-aliasing resolve, and a temporal filter. As of summer 2013, TXAA is only supported on Kepler-based GeForce cards, such as the GTX 600 Series. 2xTXAA looks about the same as 8xMSAA and performs about the same as 2xMSAA. 4xTXAA looks better than 8xMSAA and performs about the same as 4xMSAA.
SMAA	Enhanced Subpixel Morphological Anti-aliasing Combines MLAA with MSAA or SSAA and adds a number of features that improves image quality, such as: More reliable edge detection. Better preservation of sharp geometric features and diagonal lines. Accelerated and more accurate pattern classification. SMAA is an open source project and free software is available that will add SMAA to any game.

Anti-aliasing modes

In multi-sampling AA methods, multiple samples are calculated for each pixel on the screen. This table lists the resources from which samples are made and the number of samples, per pixel, that are calculated for each resource.

• S: Supersampling. Modes ending with S include super-sampling. They also use Rotated Grid MSAA.
• Q: Quality. Modes ending with Q double the number of MSAA samples as compared to their regular counterparts.
• G: Gaussian. Applies to TrAA modes.

Mode	Samples						Alternate name
	Shader	Texture	Color	Depth	Stencil	Coverage
No AA	1	1	1	1	1	0
2xSSAA	2	2	2	2	2	0
4xSSAA	4	4	4	4	4	0
8xSSAA	8	8	8	8	8	0
2xMSAA	1	1	2	2	2	0	2x
4xMSAA	1	1	4	4	4	0	4x
4xMSAA + 2xSSAA	2	2	4	4	4	0	4xS
8xMSAA	1	1	8	8	8	0
16xMSAA	1	1	16	16	16	0
4xCSAA	1	1	4	4	4	4
8xCSAA	1	1	4	4	4	8
16xCSAA	1	1	4	4	4	16
32xCSAA	1	1				32
4xMSAA + 8xCSAA	1	1	4	4	4	8	8x
8xMSAA + 2xSSAA	2	2	8	8	8	0	8xS
8xMSAA + 8xCSAA	1	1	8	8	8	8	8xQ
4xMSAA + 16xCSAA	1	1	4	4	4	16	16x
16xMSAA + 4xSSAA	4	4	16	16	16	0	16xS
8xMSAA + 16xCSAA	1	1	8	8	8	16	16xQ
8xMSAA + 32xCSAA	1	1	8	8	8	32	32x
32xMSAA + 4xSSAA	4	4	32	32	32	0	32xS
4xMSAA + 4xTrAA	?
4xMSAA + 2xTrAA	?
OGSSAA + RGMSAA	?
							4xG
							2xQ
							6xS

Note

Many of the terms are coined by NVIDIA and describes AA modes implemented in their hardware. Some terms are coined by ATI and some are generic.

Note

The sample counts for the CSAA modes are inflated due to NVIDIA counting Multi-samples twice, once as Multi-samples and once as Coverage samples.

Note

The sample counts in the more esoteric AA combinations have not been verified.