Originally Posted by scupkin
I just manual set my ram from auto 5,5,5,15 667MHz to (1:1FSB) 266MHz 4,4,4,12. The lower the latency the faster it should be right?
Lower latencies are faster at a given clock speed than higher latencies, but higher latencies allow higher clocks. Usually, the best method is to run high latencies to find you maximum ideal overclock. If your maximum overclock is not RAM limited, then decrease the latencies to gain a little extra performance.
Use sisoft sandra to test raw ram bandwidth to find out whether a 1:1 RAM divider is faster than the 666 MHz divider. Some chipsets don't mind so much. Others, like mine, actually perform worse with the asyncronous divider.
In Sandra, my system made 2900 MB/s with a 1:1 divider, 574 (overclocked) FSB, 4-4-4-10 timings. With the divider set to run the RAM at 667 speeds, the throughput decreased to 2600 MB/s running the same 4-4-4-10 timings, even though the RAM was technically running faster.
So I stayed with the 1:1 divider. Since the clock speed is nice and low, I am able to run the timings at 3-3-3-8 and drop the command rate to 1T. This bumped my Sandra score to 3000 MB/s. Clearly the best choice! Hope this clears things up.