In order to check the performance impact of the high-speed memory, we ran tests using four configurations, using quad-channel and dual-channel (installing two modules at the first and second channels, and leaving the third and fourth ones unused) modes, at 3,000 MHz and 2,150 MHz.
But why did we tested at this speed? The point is that, when you select the XMP setting for 3,000 MHz, the motherboard sets the base clock to 125 MHz, and not 100 MHz as usual for X99 systems. The memory clock multiplier is set to 12x, which brings a 1,500 Mhz clock, that gives the 3,000 MHz effetive memory speed.
In X99 systems, all the other clocks can be left on their default clocks, even when you raise the base clock. But you have to adjust your CPU multiplier to keep your CPU running at the desired clock. So, we set the CPU clock at 3.75 GHz for the test with the memory at 3,000 MHz. If we just used the 2,133 MHz or 2,400 MHz standard memory settings, the CPU parameters would change, which could bias the benchmark. That’s why we kept the base clock at 125 MHz and lower the memory multiplier, so it was working at 2,125 MHz, leaving all other configurations untouched.
Figure 5 shows the SPD data, read by CPU-Z.
Figure 5: memory settings
We ran the memory benchmarks using AIDA64, an excellent tool for system information and benchmark. We ran Memory Read, Memory Write, and Memory Copy benchmarks. The results are shown in the graphs below. The values are in MiB/s.
Here, we see that, in quad-channel mode, the higher (3,000 MHz) speed brings no performance increase; this is due to the limitations of the CPU memory controller.
However, in the dual channel benchmarks, we measured an increase of the bandwidth, up to 31%, using DDR4-3000 memory instead of DDR4-2150.