ATI Radeon HD 2900 XT test

Unified shaders

Before we look at the new ATI chip in depth we will firstly look at the unified shader phenomonon. It is the most important difference between DirectX 10 and DirectX 9, and we will take a short trip back in time to try and explain the exact workings of the new technology.

Up to 2001 3D graphics cards worked to a predifined set op rules and could only show a limited amount of 3D effects. When DirectX 8 was introduced this all changed and developers could start to program their own unique 3D effects. To aid these effects the vertex shader units and pixel shaders units were introduced. A vertex shader effect is able to influence the postition of the corners (verticles) of triangular objects. Pixel shader algorythms allow a developer to execute effects to a pixel level, which means that every pixel on the screen could have a different color or shade.

Because a GPU is split into seperate vertex and pixel shaders the current GPU's are hardly fully used to their potential.

In a conventional GPU the vertex and pixel shader programs are handeld by seperate units in the chip. An ATI Radeon X1950 XT for instance sports a total of 8 vertex shaders and 48 pixel shaders. This does mean that the full potential of the card will seldom be reached playing 3D games, as some scenes will rely very heavily on the vertex shader and leave out the pixel shaders or the other way around. The concept of unified shaders introduced in DirectX 10 solves this problem.

As the name indicates a unified shader can be used for all sorts of calculations, and can act like a vertex or a pixel shader, depending on the situation or scene in a game. It also means that all shaders within a GPU can be fully utilized, beacuse when a scene requires more vertex than pixel shaders the majority of the unified shaders on the chip will handle these vertex calculations, allowing for a lot more flexibility in the programming of 3D games.