SANDRA Benchmarking software

I’ve just installed the SANDRA diagnostic and benchmarking software on a computer, to familiarize myself with its capabilities.

The RAM architecture is single channel DDR-2, with two DIMM slots.

Per instructions from my instructor (I’m doing this in a training program for work), I benchmarked the RAM Bandwidth using first, a single 512MB RAM card, and then two cards. In both instances, I got an effective bandwidth of approx. 3110 MB/s.

I then swapped out my memory cards for a pair of 256MB sticks, and ran the diagnostic again. The single-stick bandwidth was 3232 MB/s, and the two-stick bandwidth was 2950 MB/s.

This did not seem right to me, or to my instructor. Why should doubling up on the amount of RAM cause a drop in bandwidth? Postulating that one of the memory sticks might be compromised, I re-ran the diagnostic, using the stick that had not originally been tested in a single-stick configuration.

The results were the same. RAM bandwidth of 3232 MB/s.

Anybody have any ideas why this should be so?

Thanks in advance for any insights.