ABIT NI8 SLI: Memory Overclocking

Each generation of boards and chipset technology often comes with even greater opportunities to increase a board's bandwidth and performance through overclocking of the memory. Of course on Intel boards, the optimum scenario is usually one in which the FSB and memory speeds are 1:1 with the tightest timings possible. Increasing memory speed beyond the FSB speed yields not much gain, and more or less, it is left for bragging rights in the ability of a mainboard and/or CPU to push its ram into the stratosphere. In many cases, to achieve ultra high memory speed, you must reduce timings and latency to such an extent that bandwidth begins falling off considerably.

We'll start with our baseline Corsair at the previous DDR2 667 MHz 3-2-2-8 1T timings. All Sandra bandwidth is measured unbuffered.

The next step was to verify the Corsair at its advertised 733 MHz at 3-2-2-8 1T timings, at 2.1V. That was accomplished without issue. The memory was incrementally increased until we reached its top stable plateau while maintaining the advertised 3-2-2-8 1T command timings, 770 MHz @ 2.2V. This is at a FSB bus speed of 220 MHz (880 MHz) and total CPU MHz of 3520.

Click to enlarge.

The next progression upwards was over the 800 MHz mark. In order to attain 807 MHz memory speed, it was necessary to drop 1T command to 2T. This increase in latency can and does result in major drops in bandwidth that may not look substantial in bandwidth test numbers, but can reflect a big difference in gaming fps performance, for instance. The Ras to Cas delay and Ras precharge also had to be dropped a notch to 3. As you'll notice, compared to the previous example of 770 MHz using tighter timings, bandwidth has tailed off slightly by a little over 100MB/s.

ncG1vNJzZmivp6x7orrAp5utnZOde6S7zGiKoaenZJavsMSxZmpwYWiMpJzAoJx2alaWua2JpZqjrJ1WqLyzwJxpXamZl5qKd3LSpaygdQ%3D%3D