cjl :
In the ideal case (a perfect memory controller), triple channel will always have exactly the same latency and exactly 1.5x the bandwidth of dual channel with all other factors the same. It's the same idea as with graphics cards - the 384 bit bus on the G80 core should have exactly 1.5x the bandwidth and exactly the same latency of the 256 bit bus on the G92. Both should have perfect linear scaling with frequency (I.E. DDR3-2000 is exactly twice the bandwidth of DDR3-1000, regardless of interface). From the tests I've seen, there's also some architectural advantage with Intel's memory controller, where an I7 with tri channel will have >1.5x the bandwidth of an AM3 dual channel setup with the same RAM. This has nothing to do with an inherent advantage to tri channel though, and this will probably be seen when the dual channel Intel IMCs start showing up later this year.
Of course, the question then becomes whether the extra bandwidth from tri channel is really needed. In the case of most desktop applications, the answer is fairly resoundingly no.
Of course, the question then becomes whether the extra bandwidth from tri channel is really needed. In the case of most desktop applications, the answer is fairly resoundingly no.
Thanks for explaining that - makes sense now.