As part of the DIY ViDock project, it has been identified that ICH7/8/9M and HM57 I/O controllers are not running to their advertised 2.5Gbps in either direction. There is somewhere between 15-30% less performance than expected. It is not clear why these port are not seeing full performance. Is the port sharing bandwidth with another port? Is it running only half-duplex? This problem was identified by the following signs:
1. noticably slower performance compared to x1 1.0 set Nvidia MCP79 chipset as shown here.
2. full x1 1.0 speed is possible if the port is manually switched to have a link width of x2 as shown here. That can only apply to odd numbered ports with no device attached to the next sequential port. We are calling this x1E mode "Intel-Enhanced".
Point 2 then raises the question: how can *any* pci-e port be switched to gain full x1 1.0 performance? (2) above restricts the performance benefits to only some setups. I have tried numerous settings by referring to ICH9M I/O datasheet with no success so far.
If this problem cannot be corrected, then I would like to ask Intel to check their advertising of their pci-e ports as the performance is certainly not 2.5Gbps in either direction. There is precious bandwidth that needs to be uncapped.
Can anyone from Intel help with this issue? Surely there must be an undocumented PCI Configuration Register that can allow an even port to run at full x1 1.0 speed?