More AGP modes, transformation

AGP modes (cont'd)

Sideband Addressing provides a separate data path for commands to be sent across the bus. With a data signal, you can either send data one way or the other. It is impossible to send data in two directions on a single data pipeline. Thus, in a situation where data is being uploaded, it is impossible to send a command to the CPU simultaneously. Therefore, the upload must be stalled to send the command, potentially offering a significant performance reduction. Sideband Addressing provides this separate command path, allowing commands to be sent without interrupting any operations.

Fast Writes is another AGP technology that has only been implemented by NVIDIA thus far. This technology allows for direct writes to the graphics card from the CPU. For a graphics card to traditionally access data over the AGP bus, it must first be written to system memory and then accessed. With Fast Writes in the picture, data can be sent directly from the CPU to the graphics card, removing the latency associated with reading and writing data from system memory. This too can help improve performance in certain situations.

Transformations

Every scene a computer renders is made up of geometry. Triangles are the simplest form of this geometry, and they are the basic structure of every object that is rendered. However, computers don't truly render triangles, but at a lower level they use vertices. A vertex is a simple point, or position in space. Think of it as a small dot in the 3D world. Each vertex represents part of a triangle or triangles, with the most basic form of a triangle having three vertices, with one at each corner of the triangle.

When dealing with vertices, there are many mathematical calculations that must be processed in order to display them onscreen. These calculations are known as transformations. They work by performing mathematical calculations on each vertex that result in translating, rotating, and scaling them between "spaces". There are three spaces that each vertex goes through: Object, World, and Screen.

Object space is the first of the spaces and it is the base level at which the vertices exist. The next space - World - is where lighting (which will be covered in the next section) and culling take place. Screen, the final space, is where the scene can be rasterized and textured.

T&L engines can support different features that are extensions beyond the standard transformation. Higher Order Surfaces is one of these extensions, which deals with geometry compression. There are a variety of implementations for HOS, with the idea behind them using a general description of an image to create a highly detailed version of it. For example, control points can be used to define the general shape of an object, effectively providing a description of what it looks like. From there, a tesselator takes this description and sub-divides the scene into triangles until the desired level of detail is achieved. Each sub-division by the tesselator provides greater detail, typically making the object smoother and rounder. Another implementation is simply rendering a low-detail version of the object and tessellating that down using N-patches.

Vertex shaders are another extension of the T&L engine, allowing for programming specific functionality into each vertex. Doing so allows for the vertices to be manipulated achieving different effects. For example, a vertex shader can be used to achieve wave effects that allow for simulating water. This is achieved by performing specific operations on each vertex to calculate the new location. Graphics hardware that does not support vertex shaders must resort back to software T&L when shaders are enabled.