Testing Method & Test Setup
To test the overclocking capabilities of the memory we are going to use Intel’s
recently released Haswell platform. As memory overclocks are known to vary between
different motherboards, we are going to perform the tests using two different
platforms to be sure that our numbers are reliable.
Motherboard |
ASUS Maximus VI Gene (BIOS 0607)
Gigabyte Z87X-OC (BIOS F5q) |
CPU |
Intel Core i7-4770K ES @ 4.0 GHz |
Graphic
card |
ASUS GTX 580 |
Memory |
ADATA AX3U2800W4G12-DGV |
SSD |
Samsung PM840 Pro |
PSU |
Seasonic Platinum 660 Watts |
OS |
Windows 7, 64 bit SP1 |
Even though Haswell is very flexible on the memory frequency one can set, very few
people actually do base clock (BCLK) overclocking on their daily setups.
Therefore, instead of our previous procedure of fixing the voltage and raising
the frequency in 10MHz steps we are now going to fix the frequency and minimize
the voltage in 0.01V steps.
As usual, our stability method of choice is HCI Memtest.
Since we are dealing with an 8 GB kit, we use eight 750 MB instances and call
things stable if we see all of them to go past 100% without showing a single
error.
Not to get things too complicated, we only set the primary timings, command rate
(1T) and the memory voltage by hand while the rest of the settings is left for
the motherboard and SPD to agree on.
Results
Traditionally, our aim is to see how memory reacts to voltage changes and in case with Hynix MFR this reaction translates in almost linear ability to run higher frequencies without having to raise the CAS latency. What comes to second and third entries in the primary timings list, tRCD and tRP, the main factor for lowest stable values of those is frequency, with voltage playing a role only in borderline areas. Lastly, the tRAS value we use in each part of the testing is a sensible value induced by an educated guess.
As could be expected from a top Hynix product, 2800C12 version of XPG V2 shows impressive results throughout the entire testing range. For a start, our CL9 and CL10 adventures has brought us to, respectively, DDR3-2400 and DDR3-2666 being held by inability to scale past 1.8V both times around. When it came to CL11, we could get the memory to run 11-13-12-32 at its rated speed with only 0.02V voltage increase and using CL12 we saw more than 66MHz headroom above specification.
While DDR3-2933 is where the capabilities of our CPU end with respect to full stability, but it’s not the end of the whole story. Eager to see what the memory is capable of when we relax stability levels and crank up the voltage, we ended up with SuperPi 32M passes at 1400MHz 10-13-12 and 1650MHz 11-15-15.