When AMD released its new 6000 series of video cards last week, the overall message was clear to anyone of any level of computer knowledge: new! improved! inexpensive! better than Nvidia! But something that had potential to give even enthusiasts pause was one of the specific advancements AMD touted: morphological anti-aliasing (MLAA). Why was it being introduced? And what made it so special compared to other forms of anti-aliasing?
Traditional multisample anti-aliasing (MSAA) is one of the more digestible terms often thrown around when discussing video cards. It's the technique that makes the jagged edges of pixel-created graphics look more realistic. It does this by taking (or sampling) colors from nearby pixels and adjusting them to improve the transitions from one color to the next. In most 3D computer games, you have some control over the number of samples used, a setting that's often represented in terms like 1x, 2x, 4x, and so on. For all intents and purposes, the higher the number, the better your graphics will look—and the more powerful a video card you will need in order to handle it.
MLAA, which was formally introduced just last year, works a little bit differently. First, it's a post-processing filtering technique, which means it occurs outside of the usual rendering pipeline and thus is, as AMD tells us, "API and engine agnostic." Also, it uses only color data. It essentially tracks down jagged edges in images by searching for stark color differences between pixels, identifies predefined patterns in those edges, and then smooths them by blending the colors around them based on what the patterns are. MLAA is generally considered to be more efficient than other forms of anti-aliasing, because it uses only what it needs, rather than wasting time computing color samples it eventually averages away.
There are some down sides to MLAA. It has some difficulty handling pixel-size features, such as at the extreme edges of busy images, and can slightly distort textures in some cases by trying to remove higher frequencies and thus stripping away detail. It's also not ideal for handling border pixels that lack extensive information about their surroundings. Slowly moving images might have identical sets of morphological shapes, causing MLAA to guess incorrectly about how they should be blended. And because MLAA does not distinguish between foreground and background in an image, it may distort text as well.
Most players of 3D games, however, are unlikely to notice many of these visual problems; in overall quality, MLAA delivers results comparable to (or perhaps, at least according to AMD, slightly faster than) regular supersampling, offering coverage similar to edge-detect custom filter anti-aliasing (CFAA), which AMD introduced in its video cards a year years back. (The big difference is that, unlike CFAA, MLAA applies to all the edges in an image.)
AMD has included MLAA in its Catalyst driver, with the most recent 10.10a hotfix, and it may be enabled within the software.