Higher MHZ for RAM not only gives higher bandwidth but also lower latency. Though obviously it can't be the only reason for 40% difference.
if the RAS/CAS settings stay equal, but usually the latency in ns stays the same (the cycle count increases), when the frequency is higher.
nicely seen for example on the kingston page:
http://www.kingston.com/hyperx/products/khx_ddr3.asp
DDR3-1866 9-11-9-27
DDR3-1600 - DDR3-1800 9-9-9-27
DDR3-1600 7-8-7-20
DDR3-1333 7-7-7-20
of course, there is some noise in this comparision, but in general if you look at DDR memory over the years, DDR1(400) ~ 3cycles, DDR2(800) ~6cycles, DDR3(1333) ~9cycles.
There are of course even DDR3-2500 modules and they have similar timmings to those DDR3-1866 modules, but they are technically the same modules, just hand selected.
While 1333 vs 1866 seems to be a different design (maybe smaller process?)
The bad bulldozer performance makes me especially sad, as Intel seems to have no pressure to progress any further with their CPUs, the newest leaked/rumors say, that IvyBridge will just have the performance of SandyBridge at a lower power level (95W->77W), and again just 6cores for consumer.
If AMD continue to focus on APUs, it might be even their advantage in the long term to have a weak CPU, as at some point they'll have enough benchmarks to show off the advantage. I wish intel would view IGPs as competition for their CPUs and hurry up speeding up the vector units.
Has anyone found benchmarks of the AVX units, especially FMA4, from an independent source/reviewer? I'm very curious how the single thread performance is and how good two threads running it will scale.
