Why is AA in video games so damanding?
In: 3
Anti-aliasing renders the picture at higher resolution to find the fraction of a pixel covered by a polygon. When you read 8x AA that means each pixel is divided into 8 and most calculations are done that many times instead of 1.
Suppose the GPU finds that the object covers the 3 of the points. Then the color values are combined by mixing 37% of the color of the object and 63% of the background. If the object is moving, you might find that it covers 6 samples on the next frame. Now the amount of color imparted by the object is greater. The result is that motion appears smoother because the object doesn’t have to jump by a whole pixel.
The ways to do it right essentially all require looking at a given pixel multiple times. This might not sound like much, but it’s computationally similar to running a game at a higher resolution. 4x MSAA (or Multi Sample Anti Aliasing) will look at each pixel 4 times, once in each corner and take the average color of that pixel and display that to the screen. At 1080p resolution this bumps it up from 2,073,600 “real pixels” to 8,294,400 “sampled pixels”.
Even if you have a buffer to say where the harsh edges on the screen are and use that to apply your anti-aliasing algorithm, you’re still having to look at a single pixel multiple times each frame.
On top of all of that, blurring operations are always slow, because division on computers is slow. To get the average pixel color you have to take the Red, Green and Blue values all, add them up and divide them. This has to happen every single frame.
Different types of anti-aliasing have different effects on the performance. SSAA is incredibly resource-intensive because it effectively forces the GPU to render each frame had a multiple of its final resolution and then downscale it for display. This means that the number of pixels that need to be rendered goes up proportionately to the amount of anti-aliasing you choose.
There are a number of proprietary anti-alias algorithms that take shortcuts to make it faster, but in all cases your effectively asking the GPU to render more information than you will ultimately see so that it can condense it into a nicer looking image at the end
Latest Answers