What good is buying expensive memory kits if you do not push the envelope. Crucial has been making the Ballistix brand of memory modules for many years, always with the eye towards performance nuts and overclockers.
For overclock testing we did not push the limits of stability, nor did we run any tests to determine what is “24/7” stable. Pshaw! Speeds were attempted at a given timing and voltage layout to get windows stability, then benchmarks were run until all tests were capable of passing. Which was generally PCMark or SuperPi 32M. Surprisingly PCMark full suite required the largest drop in overclock to achieve stability. All testing was performed at lowest stable speed.
Overclocking anything is largely a guessing game. Even with an experienced hand a new product is going to be unknown until you start tweaking and finding what works and doesn’t. With limited BUS clocking capability on Sandy Bridge it was obvious going much past 1950 MHZ was going to be hard to do. In truth BCLKing on Sandy Bridge is largely a hit or miss operation as without significant amounts of time it can be difficult to determine, if a particular component is failing or if the motherboard or CPU just does not like BUS clocks.
So start off we tried the stock multiplier and easily ran a SuperPi 32M at 1950 MHZ 9-9-9 with only a small Vdimm increase. Hitting the board’s post limit at about 104 BCLK.
So instead of going faster, lets try dropping it down to 1600 MHz and see if we can run at 8-8-8-24-1t. This is about what chould be expected of 9-9-9 1866 MHz DDR3, but these are 8 GB ram modules, so might not be capable of running tighter timings.
Well that worked. After BCLK at 1600 MHz, and quickly hitting the boards limit, we pushed Vdimm up to 1.6v dropped BCLK back to 100 and increased memory multiplier to 1866. But decided to just go for broke… and start climbing the BCLK as well. With Spi 32M being run ever increase.
While this proved to be stable enough for 32M, dropping it to 1900 MHz was necessary to run the entire array of tests. Specifically passing full suites of both PCMark tests.
Well surely if it can do that at 9-9-9 it might be able to go higher. Here we hit a wall. Once again though it was a problem with the motherboard and not the memory. Adjusting the command rate manually and setting Vdimm to 1.65 allowed us to get 2133 9-9-9 booted right up and running smoothly. After spending a few hours trying to get the board sorted out only to attempt pushing the BCLK up to get another couple of Megahertz out of it just did not have any appeal and testing was performed at that level.
Amazing clocking capability out of these sticks. Two of these kits on X79 might be REALLY impressive.
For testing the capabilities of this RAM a P67 Sandy Bridge based platform was used. Consisting of a stock clocked 2600K Core Second Generation processor and the EVGA P67 FTW edition motherboard.
For testing performance we used the following software
The system tests are the predominant factor in determining memory performance, with SuperPi being an excellent gauge as well. wPrime 1.55 cares little for timings but can see a small improvement with speed. Games tests likewise will show little change based on memory performance overall. Max textures and details are used in each game test with 4X AA. A screen resolution of 1920×1080 was used to conform to most common desktop monitor size based on informal message board results.
Testing was performed at stock XMP levels, 1900 MHz 8-8-8 and 2133 9-9-9 2T.