It's not so much a direct comparison of x16 and x8 because you need to know the PCIe revision. 1.1 had a maximum bandwidth of 250MB/s per lane, 2.0 doubled that to 500MB/s per lane and 3.0 doubles that again to 1GB/s per lane. Thus we have the following: PCIe 1.1 x16 - 4GB/s PCIe 1.1 x8 - 2GB/s PCIe 2.0 x16 - 8GB/s PCIe 2.0 x8 - 4GB/s PCIe 3.0 x16 - 16GB/s PCIe 3.0 x8 - 8GB/s
PCIe bandwidth really comes into play for GPU computing or 3-way and 4-way video card setups. With just 1 or 2 video cards, you won't notice a difference.
Interesting. The percentage differences on the graphics scores are probably within the acceptable margin of error for the test. Basically further proving x8 is all you really need except in specialty setups. What is intriguing is the CPU difference. That's a noticeable difference. I can't think of any reason why PCIe config of ANY kind would make such a difference.
It's possible heat played a role in the test as you undoubtedly had airflow differences but I can't imagine it got hot enough that quick to make that drastic of a change. If heat did cause that kind of a difference in what I would assume to be a relatively short test (long enough to swap the video card around, test it and swap back?), I'd say you might want to look at your cooling setup and see if there are any problems (dust filled heatsink or fan perhaps?).