Test Setup

As we continue to improve our testing stations, we have now moved on from the ATI X800 based graphics card to a more recent Radeon 4350. This move was done to allow for the use of the newest driver set and in combination with the upgrade from Windows XP to Windows 7. Using the newest operating system was a necessary step to ensure the best use of the SSD drive as well.

Performance & Overclocking

Before getting into the grinding part of this review, a quick view of the SPD programming reveals three JEDEC profiles so that the memory can properly boot on various systems, along with a clean XMP one for easy configuration of the memory.

A lot has changed since the first overclocking DDR3 hit our test bench. Not only has our testing plattform evolved with time, but the memory - in combination with specification restriction has gone a completely different way than DDR2. With the second generation DDR, almost every overclocking kit required much more voltage than intended by the JEDEC standard. It was also a good bet, that if you applied more voltage, then you would get a better overclock. This is simply no longer true with DDR3. In the very beginning the modules required a lot of voltage to attain speeds faster than 1066 or 1333 MHz and everything seemed to go in the same direction as DDR2, but Intel put a stop to that with their 1.65 V specification. We have seen memory mature differently, suddenly becoming less scalable with voltage. I am writing all this, as the Xtreem Dark kit is the first to scale worse with voltage. This may not be a bad thing for the normal user and mainstream overclocker, but does not bode well for hardcore and professional OCers.

Starting out at the intended speed, I simply set the ASUS P55 board to the XMP profile which worked flawlessly and automatically keeps the CPU at default level, while setting the divider to 2:12, resulting in the 1600 MHz speed. From there I tried pushing the memory further and managed a healthy overclock of 1746 MHz. But beyond the system became unstable, no matter how much voltage I applied. Benchmarking was still possible at that speed with 1.7 V but the system became very unstable at 1.9 or 2.0 V.

The next step meant tightening the timings to see if the memory would boot at such settings. Starting out with CL5-5-5-15, the kit did not boot - no matter how much voltage was applied. Loosening the latency to CL6-6-6-18 yielded a sucessful boot at 1.6 V at which point it was not stable in Windows and I was not able to run through the benchmarking suite. With a slight nudge to 1.65 V everything worked fine and I still had a tad more headroom managing 1342 MHz. Raising the voltage to 1.7 V yielded the exact same result, but 1.8 V actually reduced the overall headroom of the memory by 20 MHz. Setting the system to 1.9 V or 2.0 V made things even worse, as the test bench would not boot or - if the voltage was set in Windows - freeze immediately at 1333 MHz. At this point it becomes clear that the memory ICs are not meant for high-voltage. This should not surprise anyone as there is simply no reason for any manufacturer to offer kits which require more than 1.65 V anymore as everyone is moving to the new i3, i5 or i7 systems.

Well then, continuing to relax the latency to 7-7-7-21 the memory managed 1624 MHz at any voltage of 1.7 V or less. This means that you could run the advertised speed of the Xtreem Dark with tighter timings if you wish. Raising the voltage all the way up to 2.0 V meant taking a hit of well over 100 MHz as far as overclockability was concerned, as the memory only manages 1504 MHz.

Last, but not least, a relaxed timing of CL9-9-9-27 also meant that the memory was able to break the 2 GHz barrier with 1.7 V, falling short with 1.65 V. At such a setting 1936 MHz was the end of the line. While the memory allowed me to set the base clock to result in 2044 MHz, before freezing during boot or right in Windows, the best stable overclock of the Xtreem Dark was 2022 MHz. Setting the memory at CL10 is of no interest by today's standards, as that is nothing out of the ordinary anymore.




As you can see, our kit did not manage to overclock better with more voltage. Anything above 1.7 V resulted in less overclockability, at times with smaller drops and sometimes with larger ones. Any result below 1333 MHz is not part of this graph, as the default base clock of 133 MHz also means that the memory should run at least at 1333 MHz if we want to keep the 2:10 divider of modern P55 based systems. But remember, your mileage may vary!