For our benchmarking testing we used the Intel Z68 platform with a 2600K processor. This particular CPU used for our testing is limited to about 105.5bclk, so to maintain a level benchmarking playing field, we chose 40×105 and 42×100 for the memory tests. In both cases the resulting CPU speed is 4.2Ghz leaving the memory clock as the difference. 2240MHz is very close to the maximum stable memory speed with this particular CPU and motherboard.
|CPU||Intel i7 2600K @ 4.2Ghz|
|Memory||G.Skill ARES 8GAB 2133MHz 9-11-10 2x4Gb|
|Graphics Card||Nvidia GTX 580|
|Hard Drive||Western Digital Velociraptor 150Gb|
|OS||Windows 7 x64 w/ SP1|
There's no problem fitting the G.Skill ARES under the huge Noctua NH-D14.
Synthetic benchmarks such as these show the best case scenarios of the effects of bandwidth increases.
SuperPi 32M is quite a memory intensive benchmark and we can see the difference higher speed memory makes here.
Real world tests
These tests give a better indication of the effects of bandwidth increases using real workloads.
X264 is a high quality, commonly used video encoding software library. This benchmark provides a good indication of the effects of multithreading, CPU clocks and memory clocks etc. A user spending hours doing video encoding will want good fast memory to trim a little time off of their encodes.
Winrar is a file archiving and data compression program similar to Winzip and 7zip. It is particularly sensitive to memory bandwidth and latency improvements.
For a gaming test we used the now aged Far Cry 2 benchmark. This test was run at medium settings at 1920×1080 and so begins to run into CPU limitations at high FPS levels. Modern games tend to be GPU limited, so memory bandwidth improvements typically don't show any discernable effect outside of helping with minimum FPS and boosting games that are CPU limited. This is what we see here.
These benchmarks do show some performance benefits. Overclocking the memory doesn't provide massive performance gains in general usage, but these days with memory prices dropping through the floor, a high spec kit such as the G.Skill ARES 2133MHz kit only costs a few dollars more than a common 1600MHz kit. Why not drop the extra few dollars on some fast RAM?