Elsa Gladiac 921DVI GeForce3 Ti500

Author : Wayne Date : 30th December 2001

3DVelocity would like to thank Elsa for their help and courtesy in providing this graphics card for review.

Technology Overview :

Pixel and Vertex shaders

The nfiniteFX Programmable pixel shader :

In the past, graphics cards were manufactured with a set of fixed function special effects that developers could use to enhance the look and feel of their games. Because of this many games of the day were very "samey" with similar lens flare effects, similar lighting, similar texture handling and so on. In those days when a graphics card manufacturer had an idea for a great new effect you had to wait for their next card to see it in action, then hope somebody created a game that actually bothered using it. With the advent of the programmable pixel shader all that has changed. Now it's the developer who can decide the kind of effects he/she'd like to see in their new title, and then set about teaching the card how to create them. Essentially the programmable pixel shader lets a developer get to the very heart of the hardware and define a bewildering array of special techniques and functions at will.

The nfiniteFX Programmable vertex shader :

The vertex shader works in very much the same way as the pixel shader, but rather than working on individual pixels it works on vertex values and data. A vertex is simply the "corner" of a polygon, the point at which 2 of its sides meet. Using the vertex shader a massive range of effects can me conjured up including fog, smoke and other environmental effects. Other effects include heat haze, procedural deformations for creating things like soap bubbles, waving flags and the like, custom lighting effects, lens effects, motion blur and a whole lot more. Using the vertex shader you can also create morphing effects and and keyframe animation (or keyframe interpolation as ATi like to call it) and this is just scraping the surface.

The two main features added to the Ti series are 3d Textures and Shadow buffers.

3D Textures :

As a rule when something is rendered, a 2 dimensional (flat) texture is draped over a framework of polygons. What this creates is a hollow shell rather like an Easter egg. Now this is fine in most instances, but what if part of that shell needs to be removed? If you slice the top off an Easter egg you see nothing but empty space inside and the same applies to traditionally rendered objects. To accurately model an object that may need parts removing you need to make that object completely solid right the way through and this is the principle behind 3D textures. Rather than drape a 2D texture over a framework, a 3D texture is used which holds information about the internal parts of the object that aren't normally seen. By doing this, parts of the object can be stripped away in real-time without any additional calculations to quickly try to fill the empty space that remains.

Of course 3D textures are huge. In fact a 256x256x256 texture would take 64MB uncompressed, while a 512x512x512 texture would consume 512MB uncompressed!. NVIDIA counter this by uniquely compressing the texture three dimensionally and features a multiple compression ratio system that includes 1:4 and at the extreme 1:8 compression.

We also need to think about object distance. Just as MIP maps are used to render an appropriate 2D texture depending on object distance and size, so a similar routine needs to be carried out with 3D textures. By implementing Quad-linear filtering, NVIDIA are able to offer fast, manageable high quality 3D textures. Uses of 3D textures include :
Procedural Textures and Noise : For explosion effects etc.
Depth of Field Effects : To simulate blur based on distance from point of focus
Imposters : When multiple models are fixed in space, only one is rendered fully while the others are rendered using a single 2D texture to "simulate" the fully rendered model.
Volumetric Fog : Fog effects that can vary in density, width, height and even depth.

Shadow Buffers :

Shadow effects are nothing new but the GF3 Ti sets aside part of its texture unit for use as a fully featured shadow buffer. The range of effects and routines is extensive, so here's a canned look at the principle :

Firstly the scene is examined from the point of view of the light source.

From this viewpoint a shadow map is created. This shadow map is stored in the shadow buffer until called for.

The final step is to incorporate the shadow map into the final rendered scene. Using a proprietary shadow algorithm it is determined if the entire pixel is in shadow. If only part of a pixel is in shadow and the rest is lit, the precise portion of the pixel in shadow is calculated and one of 256 densities applied. This means shadow edges automatically appear soft and natural.

The great advantage with this technique is that complex scenes can be shadowed as effectively as simple ones, and to add to the realism objects are also able to "self shadow" where part of an object casts shadow on other parts of itself , such as a a tree which may cast shadows of its own branches onto its trunk.

Page 6- The card