Thanks to the modular, scalable design of PCI-Express, consisting of independent data-paths called 'lanes', motherboard designers can control a limited PCI-Express lane budget, and offer multiple PCI-Express x16 slots. In mid-range motherboards, and particularly in most motherboards based on Intel's newest P55 chipset, 16 PCI-Express lanes are used to drive two PCI-Express slots with bandwidths of 8 lanes each. These two slots can then accommodate two graphics cards for multi-GPU performance upscaling using ATI CrossfireX or NVIDIA SLI technologies. It should also be noted that PCI-Express 2.0 doubles the bandwidth available per lane. So if you read about PCI-E x8 2.0 in this review, these results are representative of a PCI-E x16 link in 1.1 mode - like on many old motherboard. Another important point to make is that PCI-Express is forward and backwards compatible. You can run any PCI-E 1.x card in a 2.0 slot or any 2.0 card in any 1.x slot, with reduced bandwidth of course.
In this review, we test the impact of running the NVIDIA GeForce GTX 480 graphics accelerator on PCI-Express slots that are electrically PCI-Express 2.0 x16, x8 and x4. We wanted to the PCI-Express x1 too, but the card did not boot up in this configuration. The lane change is done by covering the data-paths using thin, insulating adhesive tape, which turns off those data-paths (lanes), while still keeping the device functional. Tests at PCI-Express 2.0 x8 will show you how the GTX 480 with its seemingly heavy system bandwidth requirement fares on systems with PCI-Express 2.0 x16 electrical x8 slots. Tests at PCI-Express 2.0 x4 will give you a rough idea of how it fares on systems with PCI-Express x16 electrical x4 (there are plenty of such motherboards), and also systems with PCI-Express 1.1 x16 electrical 1.1 x8.
For your reference, we also wrote a similar article about the HD 5870, back in 2009.