Samsung 30 nm Green PC3-12800 Low Profile 1.35 V DDR3 48

Samsung 30 nm Green PC3-12800 Low Profile 1.35 V DDR3 Review

Timing and Frequency Scaling »


Overclocking the Samsung 30nm MV-3V4G3 sticks proved very rewarding. With only adjustments to the voltage and secondary timings, I easily reached 2400 MHz with 11-11-11-28-1T timings. Full stability did, however, require 1.575 V, which is a fair bit outside of what I would consider the "comfort zone" for these sticks. Adding voltage did not offer much in the way of frequency increases, as we easily passed 2000 MHz with the stock 1.35 V with default primary timings, and any overclock with adjusted primary timings met with very miniscule gains from 1.35 V to 1.55 V, with an average of about 25 MHz for each 0.1 V of additional voltage supplied. With the default 11-11-11-28 primary timings, we saw near 100 MHz increases with each 0.1 V additional applied to the sticks. A loss of 75 MHz in frequency because of timing changes is something that just cannot be ignored.

An overclocking tidbit or two for those with these sticks:

Timing programming:

There are two timings that are linked together that will help greatly in overclocking. When these two timings are not set correctly, overclocking will not be as successful, and long-term stability can be hard to attain without excessive voltage. The above picture shows the primary and secondary timings we used to reach 2400 MHz with 1.575 V.

The two crucial timings, of course, are CAS Latency, and Write CAS Latency.

For CAS 6 and 7, tWCL should be CAS -1 (so CAS 6 gets tWCL 5 and CAS 7 gets tWCL 6)

For CAS 8 and 9, tWCL should be CAS -2 (so CAS 8 gets tWCL 6 and CAS 9 gets tWCL 7)

For CAS 10 and 11, tWCL should be CAS -3 (so CAS 10 gets tWCL 7, and CAS 11 gets tWCL 8)

I mention this as tWCL is not a timing that is talked about very often, if at all. However, because it has proven itself as a critical timing during testing, it is something that I felt must mention.

Voltage Scaling:

The 30 nm memory ICs are manufactured by Samsung, commonly referred to as "D-die" and have two voltage modes, 1.35 V, and 1.5 V. The two different voltage modes and supporting DIMMs do use the same memory IC, so attempting to use the higher voltage setting on 1.35 V sticks maybe provides some success when overclocking, but seemingly the low-profile PCB prefers 1.35 V, and the "regular" PCB prefers 1.5 V. Both types can be purchased currently at very low prices.

The 1.35 V and 1.5 V settings each have pre-specified limits, with 1.35 V supporting 1.28 V - 1.475 V, while the 1.5 V mode supports 1.45 V - 1.575 V. I have found that the 1.35 V mode does operate at lower voltages as well, but going lower than 1.25 V can cause artifacting on the desktop with certain VGAs, even though stability testing passes with flying colors.

That said, I was able to run the default 11-11-11-28-1T timings @ 1600 MHz with just 1.2 V, which is considerably lower than the stock 1.35 V.

I also found that adjusting voltage outside of the pre-programmed voltage mode results in frequency scaling dropping off quite quickly as explained above, so I recommend that users do not exceed the pre-determined voltage ranges in order to prolong the DIMM's life. That means that 1.35V sticks should see no more than 1.475 V, and 1.5 V sticks should see no more than 1.575 V. Both types of DIMMs do seem to scale a little bit once those values have been exceeded, but because the gains are quite minimal, it is not recommended at this time.

Overclocked Performance Summary

SuperPi showed to be nearly 19 seconds faster using the overclocked settings, and the same CPU speed.

wPrime got a decent bopost as well, nearly seven and a half seconds.

AIDA Read Performance got close to a 4000 MB/s boost, and surpassed the G.Skill kit at the same speed.

Latency improved by 12.5 ns, a huge boost to performance.

WinRAR also got a good boost, again outclassing the G.Skill kit.

Shogun 2 finally showed decent gains, but did not allow the Samsung MV-3V4G3 to really beat out the G.Skill kit.
Next Page »Timing and Frequency Scaling