If I leave my FSB/NorthBridge at 1600MT/s (up from 1066MT/s) but reduce the CPU itself to 2.4GHz (via the multiplier), I still get 5.7GB/s. So that’s where the performance seems to come from. It looks like the key to high bandwidth on P45 is FSB frequency. (A few cheap Intel chips are 1066, most are 1333, and some Xeons are 1600.)
Do we yet have any indication of how core i7 is going to change things, I gather the memory controller is totally different, fsb isn’t part of the vocab anymore…
[url=“Nehalem by the numbers: The Ars review | Ars Technica”]http://arstechnica.com/reviews/hardware/ne...ch-review.ars/4[/url] Seems to suggest the memory bandwidth is going up quite a bit :)
The FSB on P45 seems to determine the actual chipset frequency. (The FSB itself stays idle during memory-gpu DMA.) A different chipset may not have this dependancy.
I have tried rearranging my RAM sticks. I’ve placed them in what is described by the P5Q Pro manual as a dual channel setting (and diagnostic software reports they are indeed working in dual channel) and in a single channel setting (reported as single channel). In both cases I had the same bandwidth both in RAM benchmarks and in bandwidthTest.
That’s inexplicable.
I’ve had some time to toy with my computer and I started overclocking the FSB. The transfers were:
@ 800Mhz (stock) - 3.1 GB/s
880 - 3.7
960 - 4.0
1000 - 4.2
The FSB definitely has much to do with the transfer I’m getting. Thanks alex, you got me on the right track.