ON GameFAQs:
Is it OK to lay my Wii down on its side?
- Theme:
- Light
- Dark
 |
 There's been much ado about Intel's forthcoming Accelerated Graphics Port (AGP). Promising anywhere from 2X to 4X increases in bandwidth over the current PCI bus, Games using 3D graphics will actually "break" the PCI bus in the near future, necessitating a new architecture. Intel's response: AGP. And while some initial AGP implementations by 3D chip makers will closely resemble PCI version 2.1, there are important differences as to how AGP will allow game developers to use more textures and scale up polygon counts in their game scenes, all the while maintaining at least 30fps. To get a better sense of how AGP works, let's first take a look at how PCI works in 3D games, and then what AGP does to speed things along. 
Graphics accelerators currently live on the PCI bus, a 32-bit bus running at 33MHz, which delivers a peak bandwidth of 132MB per second. Real-world sustained data rates are closer to the 90-100MB per second range, and while that may seem fast, demanding 3D games will soon be able to consume all of that bandwidth, and then some.
Game developers use a graphics board's video memory, also called local video memory to store two screens' worth of data, called the front buffer and back buffer (hence the 
term "double-buffering") and a z-buffer for storing depth information. The remaining local video memory, called a texture cache, is used to store textures. Any texture data the developer can't fit into local video memory is stored in system memory, and when needed, is copied into local video memory for processing. But the penalty is that the texture data is copied over at only about 100MB/sec, whereas the graphics processor can fetch data at nearly eight times that rate from its local video memory.
|