Single, dual and quad channel memory performance: More lanes, more speed?

New insights in the role of multichannel memory

By


Test

To answer all these question, we tested various memory configurations on two test platforms: a Z270 Kaby Lake and a X99 Haswell-E system. We tested the first system with single and dual channel memory, whereas we checked the performance improvement of quad channel compared to dual channel with the Haswell-E system. The Kaby Lake system was used with the IGP, while the X99 system was used in combination with a GTX 1060.

The Z270 system consisted of an Intel Pentium G4620 (of the Kaby Lake generation) and an ASUS Maximum IX Formula. This system is combined with 4, or 8 GB DDR4-2400 CL13 memory for the dual channel test. The Pentium features the Intel HD Graphics 630 IGP, which is the faster processor of the two IGPs from the Kaby Lake/Skylake generation. Pentiums from the Kaby Lake generation also come with HyperThreading, so this processor can be considered as a Core i3 without AVX2 and with 1 MB less cache (which does not matter much in practice). This system will also give a good indication of laptop performance, since dual-core processors with HyperThreading, under various names, are commonplace in laptops.

The high-end system consists of an Intel Core i7 5960X, the 8-core processor of the Haswell-E generation, and an MSI X99S Gaming 9 motherboard. The processor was overclocked to 4 GHz on all cores. The graphics card we used was the Nvidia GeForce GTX 1060 Founders Edition. We installed 8, or 16 GB of DDR4 memory, that ran at 2133 MHz with CL13 timings, for the quad channel test.

The careful reader might have noticed that we tested with different amounts of memory. Of course we verified that the memory capacity did not impact the test results: 8 GB single channel memory delivers the same performance as 4 GB single channel, etc.

We ran a bunch of tests on the two platforms. First of all, using AIDA64 we measured the bandwidth of the memory in three various tests, namely when reading, writing and copying the memory. We also ran three other tests on both systems, which were TechArp x264 5.0.1., Cinebench 15 and  and compressing files using WinRAR. The x264 encoder generally benefits from faster memory. Cinebench 15 is a good baseline for processor performance. Lastly, WinRAR is one of the applications that relies relatively heavily on the memory, at least when compression is enabled. For this test we packed and unpacked 10 GB with standard compression.

We also ran various games per platform to test gaming performance, since an IGP of course performs at a very different level compared to the GTX 1060. For this reason we tested Dota2 and Minecraft with the Pentium, which are two games that aren't very demanding graphically, which are ran quite often on IGPs. We chose Battlefield 1 and GTA V for the Haswell-E system. These two games have a relatively high CPU load, and in the past we saw that the chance of the memory having an impact on gaming performance increases when there is a CPU bottleneck. 


Also read these memory module articles on Hardware.Info

The Hardware.Info website uses cookies.
*