You need M/B and CPU that support PCI 3.0 or you won't get the best speed.
I saw 30% performance boost moving from a Core 2 Quad PCI 2.1 system to i5 PCI 3.0 system with the same graphics cards.
The data bus is just faster on newer M/B and CPUs. I see people put a cheap Celeron CPU on a decent 3.0 M/B all the time and it's just slow!
If PCIe 3.0 is twice as fast as PCIe 2.1 because of the added bandwidth, then how come I see a ZERO drop in hashrate when I mine with my GPUs plugged into the MB, using all 16x PCIe lanes, or when I'm running them off a 1x to 16x riser?
To be frank, you're not contradicting him. Your PCI-e risers still provide dedicated lanes, with the same PCI-e version
It could be so many things, including high PCI-e latency, collisions or crappy hardware attached to the bus ( integrated sound, sata, ethernet )
How am I not contradicting him? He said that PCIe 3.0 will get you higher speeds, because of the higher bandwidth.
PCIe 2.0 offers 500MB/s per lane, or 8GB/s for a 16x slot.
PCIe 3.0 offers 1GB/s per lane, or 16GB/s for a 16x slot.
So if my cards run at the same hashrate, no matter if they're running on a PCIe bandwidth of 500MB/s or 8000MB/s, why would I need to upgrade to PCIe 3.0 to run them at 1GB/s (a 3.0 1x riser cable)?
It could be PCIe latency, which is something I'm looking into lately. Collisions? Nope. PCIe isn't like the old PCI slots, where everything was parallel and shared a bus. Each lane is a dedicated line right to the controller.
