A Natural Evolution
When AGP 2.0 was introduced to replace the initial AGP 1.0 interface, it had lukewarm response as benchmarks have shown that there was little difference between the AGP2X and AGP4X transfer rate adopted by the two standards. But such technological advances normally do not happen the instant they are released as it has to work hand in hand with other hardware (in this case, the graphics card).
For a while, users have grown to accept the AGP 2.0 interface standard (normally known as AGP4X) and it has become a widely accepted standard in most of today's chipsets. However, we have depended too long on AGP4X and the transfer rates supported by the AGP 2.0 standard does seem to be a little too slow when compared with all the increasing memory, data and system buses.
AGP4X is only capable of transferring data at a maximum bandwidth of 1.06GB/s. Compared with all the existing buses in the system, it is rather low and 3D performance would bottleneck at the AGP port. There's really no point in having kickass technology like HyperTransport if the AGP is limiting data transfers at 266MHz (4 x 66MHz). Now that AGP bandwidth is doubled to 8X, the clock frequency will increase to 533MHz which means that it will transfer at about 2.1GB/s. This increase in bandwidth will directly help to improve the overall performance of the graphics subsystem. By how much? We'll investigate this further in this article.
Currently, there are not many chipsets that support this new AGP 3.0 interface. One of the first manufacturer to demonstrate AGP8X was SiS with their SiS648 chipset along with the all-new Xabre400 GPU. Other chipsets which will have support for AGP8X transfers include NVIDIA's nForce2 and VIA's P4X333/P4X400 and KT400.
Upcoming chipsets that will support the new AGP 3.0 interface standard.