All of the configuration problems mentioned on the previous page compel us to find a third Haswell-E sample just to test this memory. Geil relies on MSI and Gigabyte “cheating” with their voltage levels in order to reach a setting that’s unstable on two of our three CPUs. Ragged edge? Probably.
Getting back to the Asus board, we see that Geil’s Super Luce continues to provide moderately tight timings at lower data rates, matching the best setting of two DDR4-3000 competitors. But what about clock speed?
Because the Asus X99-Pro automatically loosens timings at higher data rates, clocking down from DDR4-3333 got us a higher ultimate data rate than clocking up from DDR4-3000. You’ll quickly notice that this is still only DDR4-3231 on a DDR4-3400 set, and it’s on the “wrong” board, but there’s other news from the MSI X99S XPower AC: even though the RAM was bootable at DDR4-3400, it couldn’t survive our Prime95 stress test at that frequency. In fact, it had to be clocked all the way down to DDR4-3227 before the errors could be eliminated. Maybe Prime95 is too stressful?
That means we had to either forgo “specified data rate” tests on the Super Luce DDR4-3400, or use the MSI motherboard to test it. And, to produce a matching CPU clock, we even had to set it to DDR4-3333. Furthermore, since MSI’s memory performance is slightly lower than Asus’, the “rated timing” performance we report will smell worse than a three-day-old pike.
For example, the DDR4-3000-optimized values look similar for all four memory kits, at 52GB/s when installed on the Asus X99 Pro. Not shown is that the same settings produced only 49.2GB/s on the X99S XPower AC. Thus, the “50GB/s” shown for DDR4-3333 should have been at least 52.2 GB/s. Such is the sacrifice of using a different board to gather those numbers.
If we ignore the “wrong board, wrong speed” XMP-based performance for the Super Luce, we see that apples-to-apples comparisons produce a lot of apples. In other words, performance levels are a dead heat in Sandra's Memory latency benchmark.
More apples, and we're still waiting for a pie. G.Skill’s tighter timings give it a slight advantage in Grid 2, while the loose timings that helped MSI’s motherboard boot with Geil’s Super Luce DDR4-3400 also hold it back in this game. We verified the motherboard problem by testing all four settings and producing results ranging from 200 to 210 FPS.
Just when we were beginning to think 3ds Max wouldn’t show a difference between various memory configurations, we suddenly find one shortfall. Unfortunately, that shortfall was accompanied by a change in motherboard, which was necessary to get the memory running less-than-stably near its rated speed.
The motherboard problem also appears in WinRAR. Apples to oranges you say? Well, this is what happens when you go to the market to buy an apple and all it has for sale is oranges.