Originally Posted by Thumper
A couple reasons, in addition to the fact that the difference in performance between 3-3-3-6 and 4-4-4-12 is pretty minimal on the Intel platform. And again, I have not tried to adjust my timings at all, they can go lower.
Note that Cwell is running an AMD system. In my testing however:
427FSB (1708) : 854 (5 - 5 - 5 - 15) is slower than 400Mhz (1600) : 800 ( 3 - 3 - 3 - 9)
This is on bandwidth, not on CPU benchmarking of course and not on system benchmarking. This is on dram throughput alone. I find that 266 (1066) : 1066 which is a 1:2 ratio shows very little difference when on the system benchmarks. A digital stopwatch loading F.E.A.R. shows little difference considering the bandwidth speed is doubled. Yes SuperPi shows a small difference but that is pure bandwidth and since there is no operations completed in ram, (in other words ram is just a repository and the only operations are completed in the Processor) this is minimal in the real world.
Each to their own I guess. I know that Nasgul and I have argued this point repeatedly. Sometimes pure canned numbers do not translate to the real world.