MaximumPC 2006 12

r&dBREAKING DOWN TECH —PRESENT AND FUTURE

64 MAXIMUMPC DECEMBER 2006

V

ideocards wouldn’t be all that interest-

ing if each new generation delivered

only faster frame rates—we also expect

them to render video and graphics that

look more realistic. Antialiasing is one tool

that videocard developers use to achieve

that objective. To fully understand anti-

aliasing, and the numerous antialiasing

techniques in use today, we fi rst need to

identify antialiasing’s prey: aliasing. And

to do that, we need to take a quick look at

display technology.

LCD is the most popular display tech-

nology in use these days, so we’ll use that

as our example; but the basic principles

are similar whether you’re talking about

CRT, OLED, or DLP. Each of these displays

creates an image using pixels; the pixels

in an LCD computer monitor are perfectly

square, with four 90-degree corners. There

are millions of them in even a moderately

high resolution display—1280x1024, for

instance—arranged in perfectly ordered

rows and columns. Each pixel consists of

three subpixels—one red, one green, and

one blue—arranged side by side by side.

Each subpixel is capable of producing 256

levels of brightness via regulated voltage

changes. Multiply 256 levels of red by 256

levels of green by 256 levels of blue and

you arrive at the now-familiar palette of

16.8 million colors.

STAIR-STEPPER

Objects consisting exclusively of 90-

degree angles, such as squares and

rectangles, look great on pixel-based

monitors because they fi t perfectly into

the display’s graph paper-like grid system.

Unfortunately, such orderly shapes are

relatively unusual in nature; and computer

displays have much more diffi culty render-

ing the curvaceous lines that are common

in the real world. Pixels are fi xed in size

and must be fi lled—you can’t illuminate

a fraction of a pixel—so curved or diago-

nal lines and spherical objects inevitably

appear to have jagged edges wherever

groups of pixels lie outside what should be

the object’s boundary.

If those edges are especially thin, they

take on a sparkly appearance; if they’re

part of an animated or moving object, they

tend to shimmer. These and similar unde-

sirable artifacts are known as aliasing, and

they’re the bane of gamers, graphic art-

ists, and animators.

Fortunately, techniques exist to combat

aliasing and cure the “jaggies” by fooling

the human eye into perceiving the edges

of digitized curves to be as smooth as the

proverbial baby’s bottom. These antialiasing

software routines are typically executed by

your videocard’s GPU.

In a nutshell, antialiasing renders more

pixels in an effort to simulate the fi ll-in effect

that smaller pixels would have on the lines.

Because the GPU can’t actually create

smaller pixels, it simulates them by altering

the shading or coloring of individual pixels

on either side of the line to create a smooth-

er, more natural appearance. If the object’s

outline is black and the background is white,

for example, antialiasing will arrange pixels

with intermediate color values along the

edge to soften the transition. The effect isn’t

perfect, as you’ll see if you zoom in on the

object in question, but it’s far better than

nothing at all.

Antialiasing in 3D graphics processors

has taken a variety of forms since 3Dfx

shook up the gaming world with the intro-

duction of full-screen antialiasing (FSAA)

at the turn of the millennium. Back then,

“supersampling” was the most common AA

technique. In supersampling, an entire frame

is rendered inside the GPU at several times

the resolution at which it will be ultimately

displayed. If the resolution is quadrupled,

for example, each displayed pixel will corre-

spond to four off-screen pixels. These extra

pixels are then sampled and colored based

on the average color values of the pixels

surrounding them. When the image is scaled

down and sent to the display, the edges

along curved lines appear smoother.

Supersampling is highly effective, but it’s

also extremely memory-bandwidth intensive.

This wasn’t a problem when typical display

resolutions were 640x480 or 800x600 and

graphics processors were expected to pro-

cess pixels numbering in the hundreds of

thousands, but even today’s most powerful

GPUs can’t perform supersampling at high

frame rates when multiple millions of pixels

are involved.

EFFECTIVE COMPROMISES

In order to reduce the crushing workload on

the videocard, engineers have developed a

more effi cient AA technique known as multi-

sampling. The key distinction between multi-

sampling and supersampling is that multisam-

pling samples only the pixels that exist along

the edges of objects, as opposed to the entire

display area. An algorithm determines how

much of each pixel resides within and outside

of the boundaries of each object and then

White Paper: Antialiasing

How videocards fi t an

analog curve into the

square digital world.

BY GORD GOBLE

HOW IT WORKS Multisampling: before and after

Because lines

that are near-vertical

or near-horizontal

can’t be drawn using

the grid of pixels that

PCs use to render

images, they fre-

quently appear stair-

stepped, as shown

here.

Using

multisampling,

the GPU renders

trouble areas at a

higher resolution, then

outputs the resultant

bits at the appropriate

resolution to render

much smoother-look-

ing lines.

ALIASED ANTIALIASED