I was wondering whether the massive bandwidth provided by a PCI-E x16 bus was necessary for block building with a graphics card, and whether x8, x4, or x1 would still work.
As setting bus lane width via software seems to be impossible, instead I cut pins 19 through 82 on a PCI-E x16 riser card. This makes the lane width the same as a PCI-E x1 connector. It would have been better to just buy a converter but this was cheaper and easier for the purpose of this test.
The nvidia driver now confirms the operating lane width as x1 instead of the x16 it was previously.
Using poclbm to build blocks, it hashes at the exact same rate as when it was operating at x16 speed, even while also displaying 1920x1080 on my monitor. Bear in mind that my card isn't exactly super new, and only hashes at around 20MH/s, but I'd speculate that even the best cards should hash at full speed using a PCI-E connector with far fewer lanes. I'm also not sure how all graphics cards handle running at x1, but I don't think it should cause any issues.
This should mean that with the appropriate connector adapters and a rather large power supply, you should be able to run as many graphics cards as your motherboard has PCI-E connectors with no loss of hash rate.
Here's pictures of the setup:http://i.imgur.com/iFe3t.jpghttp://i.imgur.com/nysoK.jpg