<< 420-D mainboards cost twice as much as the M841LR, not counting the premium added
by having to use two DDR DIMMs for full performance. >>
True, that's not in dispute.
<< And how do you get to that huge bandwidth difference? >>
No, I subtract the 2.1GB/ps of bandwidth that the Athlon's front side bus will take priority on if it's maxed out. True that the Athlon's front side bus is rarely going to consume 2.1GB/ps of memory bandwidth, but for sake of comparision, that doesn't change the fact that nForce has more avialable bandwidth. Do you see what I am saying?
<< DDR266 RAM is 2.1 GB/s per channel, and 420-D doesn't benefit as much as one might think from its dual channel setup. >>
There is a lot of miscommunication going around regarding the Dual Channel DDR and it's benefit's on nForce. nVidia went with this setup, for the sole purpose that the GF2MX in the IGP wouldn't be so bandwidth crippled like all Integrated solutions are today. And it's not about how much 420-D "benefits" from the Dual CHannel DDR setup, it's about that the GF2MX will have at least 2.1GB/ps of bandwidth with still leaving the maximum of saturation for the Athlon's front side bus. The reason however 420-D looks worse compared to an AGP MX is because the MX uses a 128-bit 166MHz SDRAM memory bus, while 420-D effectively is using a 64-bit 133MHz DDR memory bus and in the end, the AGP MX has 2.7GB/ps.
I will also comment that the reason 420-D despite the DC DDR setup, does no better than KT266A in benchmarks is because with KT266A's memory controller totally maxing out the Athlon's 266fsb, there isn't much room for improvement, another example is the KT333 launch. There's virtually no difference between KT333 and KT266A despite KT333 has 0.6GB/ps more theoretical bandwidth, why? The Athlon's fsb is maxed out as it is! I hope I came at this in a friendly manner, that was my intention.