Our stability testing methods for memory have been discussed at length in the past, so we are going to simply rely on the explanation from previous reviews. Here is a run-down of what we consider to be stable.

Memory stability, what constitutes stable? What is not considered stable? These questions get hotly debated in enthusiast forums all over the internet like little brush fires on the fringe of an inferno that play havoc with forest fire crews. Everyone has their own opinion about stability, especially when it comes to memory stability. For some, stable means they can do whatever it is on their computer without it crashing, blue-screening, or restarting; whether that means gaming or just surfing the internet. To this user, stable means simply using the computer as they normally would.

Of course, that is not what we would call stable and we do a lot more thorough testing before labeling a memory frequency/timing stable. One of the toughest types of programs on a system has been found to be distributed computing projects such as Rosetta@Home, Folding@Home, World Community Grid, and more. Running 24/7 crunching for one of these great causes is a sure way to find holes in a system if there is truly some instability, unfortunately it takes a considerable amount of time to use them for stability testing so we use the list of programs below to all but guarantee the system to be 24/7 distributed computing stable:

• Multiple loops of 3DMark 01 / 06 (30 minutes of looping the full tests each)
• Dual 32M runs of SuperPi Mod 1.5 (ran at the same time)
• 2 hours of dual Prime95 using Prime95 v25.5 Stress Testing Blend
• 2 hours of HCI MemTest Pro in Windows using all available memory

Our stability testing is most definitely up there as far as the toughest in the industry and this is because we want to make sure we are testing for the 24/7 crowd. If you are just looking for benchmarking stability then you will have to find that somewhere else, especially with this being a 4GB kit of memory. Benchmarking is generally reserved for 2x1GB kits as they are a lot easier on the memory controller and clock much higher on average. We are almost ready for the overclocking but before we get to that, we need to make sure this kit can even run the specified timings/frequency that Mushkin has outlined. This is also a great way to test compatibility on the motherboard we are using for this review, the EVGA 790i Ultra-SLI.


Specification Stability Testing:

As was just mentioned, before we start overclocking this beautiful looking kit of Mushkin Ascent memory, we want to see if it even runs the rated frequency and timings at the outlined voltage. Some readers may be wondering why that is even something we have to question but with the different chipsets available and the difficulty that brings for memory manufacturers, the stories of incompatible memory run rampant in computer forums all over the internet. The specification stability testing was added some time ago in light of this situation.

We are off to a great start because not only was this memory ready to roll out of the gate at the rated timings and frequency, but it was also quite willing to run at the lowest specified voltage by Mushkin. We simply entered the BIOS and set the primary timings to 7-7-6-18 and the voltage to 1.875v (which equates to 1.86 actual volts). All other voltages were set to the lowest possible with the exception of the SPP voltage which was running at 1.40v as we saw in the screenshot. The simple fact that we can run this kit fully stable at 1.86v should provide us with some overhead because of the voltage we have to play with going up to a specified 1.95v. Let's now see what we can make this memory due above and beyond what it is specified for.

The approach we take when overclocking is to tackle the three main timing groups that we find with DDR3 modules at the frequencies this kit can run. The stability testing has been outlined above and we will search for our maximum frequencies at CL6, CL7, and CL8. The secondary timings will be determined during testing based on how the memory responds and what it likes, let's see how high the Ascent modules are able to climb on our NVIDIA 790i platform.

Now this is how a kit of memory is suppose to overclock...and a 2x2GB kit none the less. Needless to say, we were quite impressed with what we managed to squeeze out of this memory on the EVGA 790i Ultra-SLI and these results explain why. The first hurdle was to see if DDR3-1600 was possible at CL6 and it certainly was. We had to increase voltage a slight bit over the rated voltage of 1.95v but we were able to really tighten down the timings and get the rated frequency stable at a full timing set below what these Mushkin Ascent modules are spec'd out for.

Moving up to the next common timing set of 7-7-7 provided a very nice boost in frequency but required quite a bit looser secondary timings. This may result in an actual decrease in performance as the 790i chipsets performance in memory intensive tasks really favors tight secondary timings and can actually offset higher frequencies if the secondary timings are too loose. Of course that only goes so far as raw megahertz will eventually win out.

The last timing set we looked at was 8-8-8 and again, the kit just takes off with a little bit of voltage. At 2.01v, the Mushkin Ascents elevated the stable operating frequency to a lofty 990MHz or DDR3-1980. This is just a short 10MHz away from the elusive 1000MHz mark but no matter how hard we tried, we couldn't quite get DDR3-2000 stable in Prime Blend. The one important aspect of this hefty overclock is the massive amount of SPP or north bridge voltage that is required to get full stability. Our 790i sample is modified to provide additional SPP voltage and the appropriate cooling to compensate. In the end, we required 1.67v on the north bridge to accomplish this stability which is a good pinch higher than is even selectable in the BIOS for this motherboard. You will want to keep this in mind when clocking your own kits of this wonderful memory. Below are the screenshots of the stable overclocks we just discussed complete with Memset for full disclosure of secondary timings.

We will now utilize these overclocks in the next few sections of benchmarking for a comparison of the various levels of performance of each.