interesting rdram bandwidth discovery

I always thought to get hi memory bandwidth for my th7ll it was necessary to torture my pc1066 at the highest speed possible.What i normally did was to run my 1.6a @150fsb x 4 on my memory timings. That effectively gave me a 2.4 processor speed, a 600 mhz fsb and rdram running @ pc1200. this setup gave me a sandra bm of around 36-3700. fortunately i have a real nice cpu that really cranks and this allows me to experiment with all kinds of different fsb settings and memory multipliers.on a lark the other day, i decided to crank up the fsb to 170, bump the memory multiplier to 3x and do some benchmarks. to my amazement with only a 20 mhz increase in the fsb and runnin my rdram at a modest pc1020 (a 180 mhz drop from my original setup) i got benchmarks of around 3500. i have come to the conclusion i was not comming anywhere near near using the available bandwidth of my rdram when i was running it @ pc 1200, and it is more important to saturate the fsb in order to get a more efficient thruput. this allows one to come much closer to the theroetical banwidth of the rdram at any given speed . the only drawback is that the cpu used has to be a good overclocker and allow hi fsb settings. lastly, i bet this holds true for ddr also...... flapperhead

you want to give it more bandwidth though dont you? - the more of the bandwidth it has that is used the higher the latency will be - or so I have read

Ah, the fun you can have experimenting with these RDRAM boards. I experimented with that on my board too. The highest I went was 165FSB at 1.825 Vcore. This 2.8 certainly won't be remembered for it's overclocking prowess. The Prescott is coming. And at 800 Mhz plus FSB, Rambus diehards may have to turn to the SIS R658 if the i850E can't handle it.