I'm about to buy a motherboard and I have a difficulty understanding how sharing PCIe lanes works.
I have two motherboard options:
Let's shorten them to HD3 and X-SLI, respectively, for convenience.
According to product specifications the HD3 has
- 1x PCIe x16 slot running @ x16 speed.
- 1x PCIe x16 slot running @ x4 speed and bandwidth is shared with x1 slots (PCIe 2.0).
and the X-SLI has
- 1x PCIe x16 slot running @ x16 speed.
- 1x PCIe x16 slot running @ x8 speed and bandwidth is shared with the primary x16 slot (PCIe 3.0).
If I were to install two GPUs, I would like to have my graphics processing power multiplied by two but I assume that due to lane sharing I will achieve lower efficiency when it comes to multiple GPUs (please correct me if my logic is flawed).
I've done some reading about PCI-Express and if I'm not mistaken the secondary PCIe x16 slot will have a bandwidth of 2 GB/s on the HD3 and 8 GB/s on the X-SLI.
But these results don't add up for me as the HD3 will provide 16 GB/s + 2 GB/s = 18 GB/s bandwidth and the X-SLI will provide 8 GB/s + 8 GB/s = 16 GB/s bandwidth (most probably a flow in logic, I have, there). Does this mean the HD3 will yield a better performance in the case of multiple GPUs (assuming that the two GPUs are identical and will be installed on both motherboards)?
How am I supposed to compare the total GPU performance if I were to install two GPUs on those motherboards?
How exactly is the performance affected by this sharing situation?
What should my though process be when evaluating motherboards considering this lane sharing variable?