How a Graphics Card Works - Page 10

Remember how early on in the article I said that a graphics card is a lot like a 'mini PC', having its own processor (GPU) and memory? Well, the GPU on a graphics card is optimised to employ parallel processing techniques much more extensively than your average desktop CPU (yes, even your fancy dual-chip, dual-core system). A graphics card is well suited to processing data in a parallel way, since all the vertices can be thought of as independent of one another (so, to a large extent, it doesn't matter what order you do things in).

Graphics cards have something called vertex and pixel pipelines which are built into the GPU, these allow it to execute many instructions simutaneously (an instruction for each pipeline). A card with 8 pixel and 3 vertex pipelines, for example, can independently apply shaders to 8 pixels and 3 vertices at the same time. Thus the more pixel and vertex pipelines that are present on the card, the faster it can process data (NVIDIAs current top card, the GeForce 7900GTX, has 24 pixel, and 8 vertex pipelines!).

The GeForce 7900GTX: It you had to carry a piece of computer hardware around with you, this would be it.

The positive implications on rendering performance from using pipelines are quite obvious; being able to execute several shaders at one time helps the GPU to deal with its huge, (and very time sensitive), workload more efficiently. You can clearly see this relationship between pipelines and the performance in modern graphics hardware by just looking at any of the new graphics card ranges out now. The top cards from a family simply have an increased number of pixel & vertex pipelines, with the GPU clockspeed only going up a small amount (if at all) in comparison to their slower siblings.

As well as running shaders, pipelines can also be used to process geometry data, such as the way something is stored and defined in the scene. For instance, instead of using a complex mathematical equation to create a surface, 'raw' vertex data can be sent directly to the GPU, with the output then calculated by simple pipelines all running in parallel.

<<< Previous 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13 Next >>>

Article Navigation:

• Introduction
• The Basics
• Engines, Their Drivers and A Little History
• Telling Your Buffers from Your Outputs
• Feeding the Frame Buffer
• Surface Colours and Maps
  • MIP Maps
  • Bump Mapping
• Lights, Camera, Action!
• On The Buses
  • Interfaces
• Shaders
• Pipelines
• The Rendering Process - An Overview
• Improving Image Quality - AA/AF/HDR
  • Antialiasing
  • Anisotropic Filtering
  • High Dynamic Range
• End Frame
  • Links