Network Computing is part of the Informa Tech Division of Informa PLC

This site is operated by a business or businesses owned by Informa PLC and all copyright resides with them. Informa PLC's registered office is 5 Howick Place, London SW1P 1WG. Registered in England and Wales. Number 8860726.

Analysis: AGP versus PCI-e

Forget about Crossfire and SLI for just a moment and concentrate on the root cause of the new graphics revolution -- Intel’s PCI Express (PCI-e). When last we left the Accelerated Graphics Port (or “AGP”), also an Intel creation, it had reached technical specification 3.0 and was known to most of us as AGP 8X. This final successor to AGP 1X, 2X, and 4X was officially released at the end of 2002, promising a 2.1GB/s bandwidth.

Zowie! That’s large -- or at least large enough to last for about two and a half years. Then, with the impending doom presented by the high bandwidth demands of high definition video and the constant push of gaming, it looked like AGP was running out of steam. It looked that way because it was. Shortly after the release of AGP 8X, Intel began talking about its successor, the Peripheral Component Interconnect Express (PCI-e) interface.

Advertised Specifications

PCI-e was originally called 3GIO (3rd Generation I/O). Its presence first became real in 2004, and sincere in 2005. So sincere, in fact, that it will soon kill AGP. PCI-e’s one-way bandwidth checks in at roughly 4GB/s and, if you get it going on two directions (full duplex), it will speed along at 8GB/s. While that’s anywhere from roughly 2X to 4X AGP’s bandwidth, if you really want to be wowed, think about it in terms of PCI’s original 132MB/s bandwidth.

But what do those numbers mean? At their root, they’re no more than a statistical analysis of the optimum performance under best-case conditions. Unfortunately, quantifying the reality of any speed differences between these two interfaces is not necessarily simple. An AGP motherboard and a PCI-e motherboard, no matter how similar, can never be identical. Manufacturing tolerances alone dictate that there must be a difference between the two.

  • 1