Benchmark Methodology:Now that we know what our G.Skill sample is capable of running, let's see what kind of performance we get out of it at the various memory overclocks. We have also provided DDR2 results from a system for comparison sake so let's look at the specifications of the system used in today's benchmarks: - Motherboard: Abit IP35-Pro
- Processor: Intel C2D E6850
- Processor Cooling: Thermalright Ultra 120 w/AD1212MS-A73GL 2050RPM/80.5CFM
- Memory: Buffalo FireStix 2x1GB PC2-9600
- Power Supply: Silverstone Zeus ST56ZF
- Video Card: XFX Alpha Dog 8800GTS 512MB
- Additional Fans: 120mm AD1212MS-A73GL 2050RPM/80.5CFM
- Hard Drive: 1 x Seagate 7200.9 80GB SATAII 8MB cache
- OS: Windows XP SP2 (with recent updates)
This above setup will be shown as the transparent blue bar (2) in all of the benchmark graphs. Previously we had used a DDR2-1200 comparison for our DDR3 reviews but decided to change it up a little to keep things interesting. DDR2-1000 with 4-4-4 timings is a very common overclock with a lot of memory these days and should provide some interesting comparisons to the specification results from the G.Skill DDR3 memory.
The grey bar in the graphs (1) indicates the specified frequency and timings of the G.Skill F3-12800CL7D-2GBHZ. For the grey results, the secondary timings were left to AUTO as they might be with most users and the CPU multiplier has been lowered to 8X in order to bring the CPU frequency down as close to the stock E6850 frequency without going below it as possible. This set of results is used to try and accurately portray the system settings that most users will be left with should they purchase this memory and do only what is necessary in the BIOS in order to get it running at the specified speed. Of course, when 400FSB CPUs come out, DDR3-1600 will no longer be an 'overclocked' memory speed. It will simply be another BIOS selectable option with the CPU at the default FSB and frequency.
The last three red results in the graphs (3 / 4 / 5) represent the overclocked settings that we found to be stable previously in the overclocking section. The first two red results are going to be a head to head comparison of two very formidable overclocks and should provide quite the show right through the testing. The appropriate lengths are taken to ensure an even comparison including a methodical way of installing the operating system, drivers, and programs to provide as equal a testing ground as possible for all of the settings. The testing methodology for the results we are about to see is as follows:
a/ Windows is installed using a full format.
b/ Intel Chipset drivers and accessory hardware drivers (audio, network, GPU) are installed followed by a defragment and a reboot.
c/ Programs and games are then installed followed by another defragment.
d/ Windows updates are then completed installing all available updates including .NET Framework followed by a defragment.
e/ Benchmarks are each ran three times after a clean reboot for every iteration of the benchmark, the results are then averaged.
 For all results, the same theme, and other system options are identical between the two systems as well as between the overclocked results with the G.Skill modules. The benchmark versions are all listed in the graphs and were the most up to date versions at the time of testing.
The only other settings to note are the tRD (Performance Level) and chipset strap. In the past, all benchmarks were run at the same tRD of 7 and the chipset strap remained at 266. On all the setups the chipset strap will remain at 266 including the IP-35 Pro results as that is the strap chosen on the 4:5 divider. There will, however, be an adjustment to how the tRD will be handled. In the sample graph above we can see that each result has a different tRD (noted by the PL which denotes Performance Level). This is because we realized that when running a lower FSB, for instance with the 6-6-5 timing set, most systems will likely provide a tighter tRD and thus increase performance. As a result, we were actually handicapping the lower frequency results in the past because of this. So for this review, we have decided to take the approach of using the lowest tRD that is possible for each overclock to provide a more accurate 'real-world' comparison.
We have tried to outline the testing method as best we can here but if there are any further questions about settings or the OS setup, please do not hesitate to ask in the discussion thread which is linked to at the bottom of the review. So without further delay, let's see how the benchmark witnesses will influence this case. | 
