^
Maybe under crossfire it does. AMD did admit that their crossfire connector lacks the necessary bandwidth so they have to transfer data through the PCIe bus.
Nvidia, on the other hand seems to have things well in hand with their SLI bridge and thus isn't negatively affected.
Found the AT articles you meant.
Higher link speeds does show some scaling on the 7970, but in almost all cases PCIe 3.0 x4 (equal to PCIe 1.1 x16) seems sufficient.
It's only when things drop down to 3.0 x2 (basically PCIe 1.1 x8) that things get nasty.
AT's
Nvidia Titan results show no difference at all for single monitor, much the same as the pugetsystem's tests above:
But does show a tiny bit of scaling with multi-monitor. Possibly something to do with frame buffers being copied around for each screen?
.: PCIe 1.1 x16 is more than sufficient.
I also recall reading another article that shows the link speed matters less when frame rate is bound by the render-side of the GPU. Can't find that link at the moment but I'll update this post if I do.
Yan wrote:Are these synthetic tests really that useless? Or is what they measure actually unrelated to the video quality I observe?
From passmark themselves:
http://www.videocardbenchmark.net/graph_notes.html Real Life Performance Comparison
The rating the Video Card’s are given here represents their peak performance for the type of load generated by the tests and will not necessarily match the real world performance with any specific software application or game.