C1060 Bandwidths

In my development system I have a C1060 and an 8800GTX.

With pinned memory, the C1060 host to/from device bandwidths sit at around 1.5Gbytes/s. The 8800GTX bandwidths are around 2.7Gbytes/s.

Anyone else seen anything like this?


Do you have the 8800GTX in a PCIe 1.0 x16 slot while the C1060 is in a 1.0 x8 slot?

Hrm. Not sure. I’ll go ask someone.

I’m using the CUDA 2.0 SDK (for some reason the 2.2 is puking on me now – but I don’t suppose this should make a big difference for simple bandwidth tests?)

I just noticed a dual 8800GTX machine doing the same thing.


No real information yet, but I suspect you could be right tmurray.

Unfortunately the C1060 and the 8800GTX can’t just be swapped :(.