Results

Performance memory is worth the extra money only when it’s used on a performance platform. These days, the name of the game is Intel Ivy Bridge, so that would be our primary testing rig.

Motherboard	Gigabyte Z77X-UD3H (BIOS F9)
CPU	Intel Core i7-3770K (ES, E0)
Graphic card	XFX 8600 GT
Memory	Patriot Memory Viper Xtreme Division 2 Edition PXD38G2133C11K
HDD	Samsung 40 GB
PSU	Silverstone OP1000
OS	Windows 7, 64 bit SP1

Our experience suggests that Prime 95 and HCI Memtest is the couple of tests feared most by unstable memory. We’ll select eight 750MB instance 100% pass of the latter as our yardstick for stability, as it is relatively quick and familiar for us way of testing things.
Even though Ivy Bridge has brought a lot more flexibility to memory overclocking in comparison with Sandy Bridge, it still has a number grey areas which you can’t reach manouvering with available memory multipliers and adjusting the BCLK between 98 and 108MHz, the values we found out to be the extremes of full stability. Because of that, we use a more flexible secondary platform, which will also be used to verify the overclocking results which we thought made little sense on the Ivy Bridge.

Motherboard	ASUS M5A99X (BIOS 0902)
CPU	AMD FX-8150
Graphic card	XFX 8600 GT
Memory	Patriot Memory Viper Xtreme Division 2 Edition PXD38G2133C11K
HDD	Samsung 40 GB
PSU	Silverstone OP1000
OS	Windows 7, 64 bit SP1

Unfortunately, due to memory controller limitations, which regular users are also very likely to encounter, both our platforms cannot support stable memory operation at frequency in excess of 1250MHz or DDR3-2500 so no testing above that will be done.

Twenty timing times five voltage settings is a lot of information to swallow at once, so we’ll analyse the results step by step.
Firstly, we see close to no scaling from voltage when using “straight” timings like 8-8-8, 9-9-9 and so on. This could be partially explained by the fact that most ICs these days have so-called “walls” using certain tRCD and tRP values which stop memory from scaling from voltage with a fixed tCAS value.
Secondly, we notice an emerged pattern that in order to maintain linear scaling from voltage at least up to 1.65V, tRCD must be equal to tCAS+2 and tRP to tCAS+1 which we highlighted in bold on the results’ table. Similar behavior is characteristic for nearly all memory chips currently available, not only Hynix 2Gbit CFR, so it explains the presence of lots of kits with “uneven” timings on the market.
Thirdly and most shockingly, we see that our sample has failed to achive rated speeds topping out at 1050MHz (DDR3-2100) using 11-11-11-30 timings regardless of voltage. Yes, we do use a tough testing procedure but so can the actual users – there is rendering software, for example, as sensitive on stability as the HCI MemTest we use. Having taken a deeper research into the issue, we found that the culprit is the 1T command rate, which both our motherboards default to. Setting it to 2T instead, as it is prescribed in the XMP, made things stable up to 1070MHz, which can be classified as a specification pass. Still, if memory is barely capable of doing rated speed, it won’t leave much headroom for the annual 2-5MHz overclockability loss caused by normal degradation, so in this aspect, Patriot are just begging for high return rates unless they tighten up the binning procedure.

Page 1 - Introduction	Page 4 - Results
Page 2 - Closer Look	Page 5 - Conclusion
Page 3 - Photo Gallery

Discuss this article in the forums