Intel quad channel memory performance
I have read through the Phoronix review titled Intel Quad-Core Memory Performance, which is the only review of quad channel memory performance that I have been able to find. On page 4 of that review (http://www.phoronix.com/scan.php?pag...item=606&num=4), there is a bar chart that shows compilation times in seconds for a Linux Kernel Compilation benchmark. This chart appears to mix together the affects of increased memory channels as well as increased memory, so it is difficult to quantify the separate benefits of quad memory channels versus single-channel and dual-channel memory configurations.
Does anyone know of a test where the number of memory channels in use is changed from single-channel to dual-channel to quad-channel, all the while keeping the total amount of memory the same? e.g. a 2 GB system consisting of 4x 512 MB modules, with its memory modules configured to work in either single-channel, dual-channel or quad-channel memory mode. By suitably arranging the memory modules, that should be possible, at the very least for dual-channel and quad-channel configurations, subject to the memory slot configuration of the motherboard.
It would be useful, at least for me, if it were possible to revisit the original article and a new set of benchmark single/dual/quad channel memory tests performed on a newer system. This could be one built around the Intel Xeon X7350 2.93 GHz processor or the Intel Xeon X5460 3.16 GHz processor. These seem to be likely candidates for server systems used for performing finite element analysis of large models, which is my area of interest. Large finite element analysis jobs require lots of RAM and high-speed memory I/O performance in order to help minimise analysis times.
At this point, the Linux Kernel Compilation benchmark appears to me to be the closest that I can find to a finite element analysis run, so it seems to be the best that I've got to go on. I'm certainly glad that the original Phoronix review was done.
i too am interrested in this, i would also be really interrested in seing various benchmarks of fbdimm vs normal ddr2 and possibly ddr3..
i wish i had the hardware to do a really comprehensive usable benchmark..