Memory Performance Results

Test System

I am using a fairly decent CPU overclock for all testing because greater CPU overlocks allow for greater memory performance increases to be utilized properly. I have picked many different benchmarks that show these differences, but not all workloads are going to see the same gains as these hand-picked benchmarks show. To show the increases, I started with two different memory kits, one from Samsung and one from Patriot. The Samsung kit is a 1600 MHz kit that features normal JEDEC timings for its speed; it is rated at 11-11-11-28-1T. The second kit is a much older high-performance kit that was meant to be used with P55 chipsets but missed out on full support with a Sandy Bridge CPU, since those CPUs didn't support the Patriot kit's 2000 MHz XMP speed with 9-10-9-27-3T timings natively. Ivy Bridge differs from Sandy Bridge in many ways, but one of the most important changes for gamers and overclockers is the addition of many more memory dividers, such as 2000 MHz, which makes this old kit useful again with Ivy Bridge as it wasn't matched to any of Sandy Bridge's default multipliers. I have spent several months testing these kits with various boards and both have worked great. I also test with a single ASUS MATRIX HD 7970 3 GB video card at an overclock of 1050 MHz for the core and 1500 MHz for the memory. This helps eliminate any sort of GPU bottleneck that might be introduced while, at the same time, showing to be sensitive to memory performance changes. Whether this is due to extra CPU or memory load is not known, or relevant. Let's take a look at what performance increases the Kingston kit offers:


Initial boot-up with XMP enabled was pretty easy, although I had to set secondary timings manually, as I have had to with many kits already. It's worth noting that the tertiary timings here are really tight due to the 1066 MHz JEDEC profile. For those that were wondering what dictates which tertiary timings are used on Z77 Express-based motherboards, it does seem to depend largely on the JEDEC profile. Some boards will scale these settings automatically as you increase frequency but my ASUS Maximus V Extreme does not. This is more an artifact of the test platform than it is of the sticks themselves. Two XMP profiles are available on these sticks: one for 2133 MHz and one for 1600 MHz. Like the Kingston HyperX Beast kit I covered a couple of weeks ago, the lower 1600 MHz profile does come with a lower voltage, which is great for heat reduction and slightly lowered performance, or CPUs with memory controllers not capable of running the 2133 MHz profile. I'm only going to focus on the 2133 MHz profile for today's testing because a quick set of benchmarks proved the 2133 MHz profile to be faster than the 1600 MHz profile. Tertiary timings, shown above in the MemTweakIt screenshot, are fairly tight, and I am keen to see their impact on performance.

I have to mention that these sticks offer 9-9-9-24 timings while running the 1600 MHz profile. It's not often we see 8 GB sticks offer such decent timings at 1600 MHz, with many kits sporting 11-11-11-27 or 10-10-10-27 timings. The wide level of compatibility at 1600 MHz and up to 2133 MHz was a real treat for me, giving me a clear idea of how these sticks would scale in my clock-scaling tests. I am very glad to see these sticks set up like this rather than having a profile that's only one bin lower than the rated speeds the sticks are sold for.

There is not much else for me to say at this point. I booted into my pre-installed OS and managed to complete all testing without much fanfare--as expected. The next couple pages have both system-oriented and 3D-oriented benchmark results for the kits. As is now the norm for me, I'll let the numbers do the talking.