Overclocking

Timing and Frequency Scaling

I also test modules for voltage requirements at different CAS settings and speeds to show how well a manufacturer has binned their modules and how well they've optimized them for the best possible performance. Would tighter timings and a lower speed make this module perform better? Can we increase or decrease voltage to get more performance? When overclocking, there are many different considerations that must be taken and this test helps with a few of those choices. For this test, I start with the default timings and a lowered voltage of 1.35 V before figuring out the maximum speed I can get. I then go through the procedure with each working CAS Latency setting and each voltage interval. Different sticks are rated for different voltages, so the range I use depends on the stock voltage and how high each voltage takes me.


Clock and CAS scaling was actually rather boring. CAS 7 would not boot at all, nor would anything lower. Neither would CAS 12, so that left me with four CAS settings to play with. The DIMMs didn't really start scaling until I cracked 1.5 V, and they seemed to scale the best with 1.65 V, but anything over 1.68 V introduced stability issues. Errors were seemingly confined to one DIMM of the four. This is part of what happens when buying kits with multiple DIMMs - they are always held back by the weakest stick. If you are interested in getting more out of a kit, binning each stick of the set individually might be possible.

Maximum Overclock

With all four DIMMs of the kit, I managed to reach a maximum overclock of 2751 MHz. This was achieved with both default timings and default voltage. Corsair Dominator Platinum modules, on the other hand, required 1.75V for stability at 2750. I was able to boot 2800 MHz at 1.65 V both by divider and by BCLK; however, 3D stability was not possible. This seems to be a CPU-imposed limit, so finding the actual maximum frequency these sticks support is not possible with the hardware I have on hand. I was also able to drop the voltage down to 1.575 V at the stock speed of 2666 MHz, so there is definitely some headroom available to this kit. I have seen users posting 3000 MHz+, although these users typically use WindowsXP, not Windows7 or Windows8. I don't know why these users prefer XP, but those traveling down the road of extreme computing are sure to discover its benefits.

With the sticks cooperating so well with overclocking while hitting a limit imposed by my CPU, I took it upon myself to tightening timings. Primary timings would not adjust any lower. However, a couple critical tertiary timings I changed hugely boosted performance. I include MemTweakIt screenshots in my memory reviews to show differences in both secondary and tertiary timings, and this is one instance where showing these differences is very important. The ASUS Maximus V Extreme has a setting called Latency Boundry that will adjust all the tertiary timings for you.


This setting features values from "1-14", with "14" being the loosest and "1" the fastest. Manually adjusting the value down to a setting of "5" was possible with the G.Skill F3-2666C11Q-16GTXD kit, but to get true stability, I had to bump it up to "6". In contrast, the Corsair Dominator Platinum sticks would only bench at a setting of "8" and required "9" for full stability. Not many stock tertiary timings changed, but the effect is to the tune of 4000 MB/s on memory copy performance, which is profound. Below are, to illustrate those differences, two screenshots taken with MemTweakIt.

STOCK:



OVERCLOCKED (with Latency Boundry setting of 6)

As you can see, all that has changed is tRRDR, tWWDR, tWWDD, and tWWSR. These settings can be adjusted for added performance while also running the stock XMP profile. That in and of itself is one of the big things making these sticks so great, since the G.Skill F3-2666C11Q-16GTXD kit manages to beat out the Corsair kit once these timing have been adjusted - even with looser primary timings.

You'll find a bunch of benchmarks that illustrate the overclocked performance boost on the following pages. They are, again, broken into System Performance and 3D Performance sections. This section includes the XMP results of the F3-2666C11Q-16GTXD kit, its overclocked numbers, and a reference JEDEC 1600 MHz set of numbers. Keep in mind that this is not the maximum these sticks can reach, since I found no need to increase the voltage to reach these clocks. In fact, I could have probably lowered the voltage a bit had I taken more time to test. As you can see, I've covered a lot of ground here, so there was only so much time left in my weekly schedule. As always, I'll let the numbers do the talking.