Making lines look
smooth isn't the only concern when it comes to how good a scene looks. Games are
more impressive when more polygons are used for the objects in them and when
hardware technologies are used to make lighting and surfaces look better. This
is where things like hardware transform and lighting, environment mapping, and
per-pixel shading come into play. Hardware transform and lighting, or T&L,
has, of course, two parts. The transform part takes over the object geometry
processing, while the lighting component handles
the various light sources in the game or application
environment. If the CPU had to do this stuff, a big performance hit
would result, but with the special hardware on the video card it's no problem.
These features allow a chip with T&L to produce
much more realistic scenes, with objects that are less blocky and more detailed
since far more polygons can be used. Systems without dedicated T&L hardware
could still theoretically render the same stuff, but the CPU would be doing all
that work, and the performance hit would be unacceptable. The drawback to
T&L is that games and applications must be specifically coded to take
advantage of its features, so older games won't benefit from it. Currently there
are hardly any games that use T&L at all. Quake 3 Arena does to some extent,
and Soldier of Fortune should benefit much more from it as apparently much of it
was coded with hardware T&L in mind. But to my knowledge those are the only
two that use it. There are numerous games in the works that will support
hardware T&L, but for the most part this feature is only something that will
make a significant difference in the months ahead.
Hardware T&L was introduced on nVidia's GeForce 256 cards, called GPUs (Graphics Processing Units) mostly by marketing people. The GeForce2 GTS's T&L unit is, not surprisingly, improved over that of the older GeForce 256's. S3's Savage 2000 chip also has hardware T&L on it, but it's inferior to that on nVidia's cards and wasn't active at all in initial driver releases. I'm pretty sure newer Savage 2000 drivers at least partially use the chip's T&L capabilities.
Several types of environment mapping techniques are
implemented in some of today's video cards.
Matrox's G400 uses environment mapped bump mapping,
which basically makes a texture look three-dimensional by altering how light
reflects off of it: it tricks the eye into thinking the surface has depth. Water
scenes benefit much from this technique, since bump making can make it look like
the water is rippling. Static surfaces, like walls, can be made to look like
they have chips or other imperfections in them. The GeForce cards support cube
environment mapping, which also deal with light reflections. Cube environment
mapping allows objects to have glossy or rough-looking surfaces. It doesn't seem
to be as flexible as environment mapped bump mapping, but it's still quite
useful for increasing realism.
The GeForce2 GTS has an extra feature that the GeForce 256 lacks: per-pixel shading and, thus, per-pixel bump mapping. As you might guess from its name, per-pixel shading handles in hardware all lighting calculations for each individual pixel. Because it handles it for each pixel, bump mapping can be achieved.
Unfortunately, these environment mapping methods, like T&L, must be supported in the game or application code, so again only programs written to use them will benefit. However, there are already several games that support environment mapping, and more are on the way.
Because they allow scenes to look so much more realistic, I
expect that all these techniques will be commonplace on video cards next year
and will hopefully be supported by most, if not all, new games.