HP Xeon 55xx GPU Bandwidth problem

Bandwidth issues for NVIDIA GPU on HP Xeon 55xx servers

But through the SDK bandwidth test, the numerical value is rather strange.

Host-> device up to 3gb/s, device-> host highest only 3.2gb/s, which is the same as the PCI Express Gen 2.0 5-6gb/s data is relatively large.

Found that other people encountered similar problems on the same HP ProLiant DL370 G6 servers.

http://forums.nvidia.com/index.php?showtopic=104243

The problem with the motherboard chip is that the chip is equivalent to integrating two Ioh, creating a strange bandwidth bottleneck.

If you want to get a relatively high bandwidth on that machine (h2d:2.8-2.9gb/s, d2h:3.2gb/s), process bindings must be made, and the test should be bound to Node 1, which is to say:

Numactl–membind=1./Running Program