I read with interest the P4 + DC DDR article. I’ve seen early Granite Bay DDR mem benches, and was not impressed.
Yes, the numbers look good compared to DDR (Ed.note: It’s about 3300Mb/sec using dual channel PC2100), but not compared to PC1066 RDRAM (I can get about 3900Mb/sec running at 151MHz FSB at 4X).
However, having an RDRAM system doesn’t limit you to running your system to around 150MHz and thus limiting your overclocking. By setting the RDRAM speed to 3X, you can put the FSB speed way up there, too.
One advantage to using the RDRAM @ 3x is the efficiency goes way up, so PC1066 using 3x, is very near or equal in bandwidth to PC1200 using 4x.
So when I use a 2.26b with a TH7-II and a mix of 2 x256 Kingston PC1066 and 2×64 Samsung PC800. I have a choice of 2550 MHz (150 x4) or 3034 MHz (178 x3). Both have about the same bandwidth, but the 3x set-up lets me increase CPU MHz by 488 MHz, and the CPU’s FSB is 712MHz vs 600MHz, hence if nothing holds you back, the 3x setting will always produce the faster machine overall. After testing 1.6a, 1.8a, 2.0a, and 2.26b chips, you need 12-15MHz more fsb to break even when switching to the 3x multiplier.
Here is my offbeat combo @ 178 x3, and compare it to your projected 3700Mb/sec for a dual PC2700 DDR system.
(The error in reporting mem capacities is not a fault with SiSoft, but with the TH7-II. This is the only downside to doing this; stability is not affected.)
The next 2 are from a Asus P4T533 @ 168 4x and 172 4x. This is not my rig, but this system uses the 32 bit PC1066 (Samsung). (Ed. note: Benchmarks published with the permission of the originator, aka “Chainbolt.”)
So RDRAM can be used for high-speed FSB overclocking, too.
(Ed. note: Again, thanks to “Chainbolt” for granting permission to use his benchmarks.)