Regular Tom's Hardware readers will remember this project, as it is not particularly new. We already analyzed PCI Express bandwidth in November 2004; that was at a time when PCI Express (PCIe) was still new, and had no significant advantage over AGP graphics. Today, almost every new computer utilizes PCI Express to hook up integrated or discrete graphics engines. Graphics solutions have advanced tremendously and we believe it's time for a new analysis, to find out the answer to this question: How much bandwidth does graphics really need?
PCI Express quickly became an important growth engine for the graphics industry, because it allowed ATI/AMD and Nvidia to deploy two and even four graphics cards into a computer. And PCI Express will remain important for additional high-bandwidth upgrades, such as physics accelerators for 3D applications and games. The additional calculation power of extra graphics cards can be used to maintain performance at increased resolutions, to add visual features, or to improve performance at common settings. This last option is not that appealing, as many of today's graphics cards are powerful enough for typical resolutions of 1024x768 and 1280x1024. The growth potential is tremendous, thanks to ATI Crossfire and Nvidia SLI dual and quad graphics, because both solutions require the user to purchase a suitable platform to support the respective standards. An all-rounder - a motherboard that supports both Crossfire and SLI - does not exist, at least not yet.
However, these dual and quad graphics setups represent only a fraction of the graphics market. Most systems and graphics card upgrades still consist of a single graphics card, which is why we did not extend our PCI Express scaling analysis to dual graphics setups either. We took typical high-end ATI and Nvidia graphics cards and ran a few benchmarks at various PCI Express link modes.