First test is to run these Corsair Vengeance LPX modules at their Xtreme Memory Profile (XMP) of 2800C16-18-18-35 with a Command Rate of 2T. To provide some other data we also included the following speeds in the charts:
- 2133C15-15-15-35 2T Command Rate
- 2400C15-15-15-35 2T Command Rate
- 2666C15-15-15-35 2T Command Rate
- 2800C16-18-18-35 2T Command Rate (VENGEANCE LPX XMP profile 1)
- 3000C15-15-15-35 2T Command Rate
Take note that the above speeds were simulated by down or overclocking the LPX sample kit and just adjusting the main timings and voltages to stabilize. No secondary nor tertiary timings were touched.
In SuperPI 32M we see decent scaling from the Intel rated 2133MHz to the whopping speeds of 2800MHz and beyond. It is no surprise that this particular Vengeance LPX kit outputs some decent scores. Nevertheless it is clear a lower specced running kit (2666MHz) with tighter timings has the edge over the Corsair XMP. So also with DDR4, tighter timings have the preference over looser timings in combination with higher memory speed. The AIDA64 Bandwidth outputs confirm our early findings: 2666C15 out edges the 2800C16 in both Copy and Read. Not by much, though the difference is there. The Write speeds don't scale at all with increased memory frequency, the solution to that can be found on the next page
The AIDA Latency scores again reinforce our findings even more. Out of the box these Vengeance LPX kit is a good performer though there is more performance in them when being run with tighter timings, albeit at a lower memory frequency for the same voltage requirement. A higher Frequency isn't always everything :)
The X264 Video encoding test and Intel's in house developed Xtreme Tuning Utility scale decently with memory speed, yet they also adore any extra bandwidth provided by setting tighter timings. As we found out during the weeks of exploring the Haswell-E platform we were kinda stumbled that the performance scaling of the memory frequency was minimalistic. In fact the X99 platform provides out of the box already more than plenty, though once we move over 2666MHz we observed the platform getting saturated and the performance scaling was getting reduced, hence why we included an extract off the Haswell-E launch article to show you what can be done to counteract this phenomenon.