Memories

Up to this point, I have focused and concentrated on getting the most out of my Intel Core 2 Duo E6700 processor using the stock HSF, except for a short detour into improving the performance of my disk drives by migrating from a non-RAID set-up to RAID 0 using the Intel ICH8R Controller and Intel Matrix Storage Console.

I am pleased with accomplishing a conservative 20% overclock of the C2D E6700 from 2.66 GHz to 3.20 GHz with a very stable system.

During this particular phase of my overclocking exercise, I intentionally kept the DRAM:FSB ratio at 1:1 to keep the memory component "out of the way" while I experimented with the CPU clock since both the CPU and memory clocks are tightly linked together. Changing the CPU clock has an immediate and direct effect on the memory clock. Keeping this ratio at 1:1 made overclocking to this point so much easier.

But is this 1:1 (DRAM:FSB) setting optimal?

Hiker, in a post in the ASUS Forums about his overclocking experience, makes a case that this may not be the optimal setting for high performance memory.

Let's find out in this next phase of my overclocking exercise. Now that I have completed getting the optimal performance from my CPU and stock HSF, I turn my focus on getting the optimal performance from my memory: 4 GB of Corsair XMS2 TWIN2X2048-8500C5 memory.

As usual, my first step was to set a baseline at my current 1:1 (DRAM:FSB) ratio using the Everest Ultimate Cache & Memory Benchmark. The baseline DDR2-640 measurements are: 6470 MB/s for Memory Read, 5832 MB/s for Memory Write, 5915 MB/s for Memory Copy and 71.3 ns memory latency.

I then changed the DRAM:FSB ratio in BIOS to 10:8 by setting the DRAM Frequency to DDR2-800 MHz. After restarting the system (with the usual automatic temporary power-off followed by a power-on which is unsettling to people who are not used to this normal Intel P965 chipset behavior), the Overclock view in Everest Ultimate confirmed the new 10:8 (DRAM:FSB) ratio with the memory clock now running at 400 MHz (or DDR2-800 MHz).

Everest Ultimate also has a "SPD" view which you can use to check the SPD (Serial Presence Detect) and EPP (Enhanced Performance Profile) settings for your memory (if it supports either or both) which are read directly from the memory sticks themselves. More importantly, you can use the information in this view to confirm that you have "matched" memory sticks, i.e., memory with the same specifications, which is important for DDR2 operation.

I used this information to verify that the memory timings in BIOS are already at the optimal settings for my Corsair memory.

So, were there any performance improvements with the memory at 10:8 (DRAM:FSB) ratio (DDR2-800 MHz)?

The Everest Ultimate Cache & Memory Benchmark showed that Memory Write stayed the same, but there was a dramatic increase in Memory Read performance and improved Memory Latency.

Memory Read performance went up from 6470 MB/s to 7491 MB/s and Memory Latency improved from 71.3 ns to 63.2 ns. The improvement in Memory Read performance directly contributed to the improvement in Memory Copy performance from 5915 MB/s to 6397 MB/s.

I also ran 3DMark06 and there was a slight improvement. You can view the comparison here.

Hiker from Sweden may be right after all.

I ran Memtest86 v1.65 overnight for over 8 hours and 38 minutes (wall time) executing 18 passes to confirm system stability with this new memory setting. No errors were detected.

The next step in my overclocking exercise is to set the Corsair memory to 12:8 (DRAM:FSB) ratio or DDR2-960 MHz. I will report on this later this coming week since I will be gone travelling on business.