XMP and near XMP benchmarks
Due to the nature of benchmarks, CPU speed is very important. The XMP profile for this ram runs the CPU about 50mhz lower then the stock speeds and 1866mhz ram speed runs the cpu, this is a small number but impacts the benchmarks significantly. For this reason I raised the CPU speed to the same 3.2ghz as the rest of the tests, and ran the ram at 2038mhz instead of 2000mhz. The nice thing about this ram is that it did it quite happily, and even allowed tightening the timings! More on that in the Overclocking section.
First, XMP speeds!
CPU-Z Information
Note that this and the following picture are as set by XMP, not after I bumped the CPU up!
All of the following benchmarks are with the ram at 2038mhz!
Everest
Sandra Pro
SiSoft Sandra scales well with memory speed, as you can see this ram is FAST!
Science Mark 2
Sciencemark needs a certain amount of ram speed to run at full speed, but doesn’t get any faster once you get past that ram speed. For this CPU at 3.2ghz that speed is about 1500mhz.
SuperPi 32m
SuperPi 32m likes ram speed and tight ram timings. It likes this ram.
Maxxmem
Maxxmem likes fast ram and tight timings as well, it too likes this ram.
XMP timings at 1866mhz
Next up, the same ram timings and cpu speed, but the ram set to 1866mhz. Because the timings are the same this isn’t overclocking, and shouldn’t pose any problems for any system. The lower ram speed allows a lower Uncore speed which makes this much more doable for CPUs that don’t like really high Uncore speeds.
CPUz
Everest Cache/Memory
We lost about 1000mb/s in all the sections, that’s not bad really, about 7%. You aren’t likely to notice that.
SiSoft Sandra
Small loses again, nothing too serious though.
Sciencemark
Sciencemark actually gained performance. This is the downside to using a largely CPU based benchmark. It does illustrate nicely how real world performance doesn’t always follow benchmarks! I’d be pretty surprised if many other problems/benchmarks gained performance at lower ram speed and the same timings.
SuperPi 32m
SuperPi 32m lost five seconds or so, not bad for an 11 minute benchmark.
Maxxmem
Maxxmem is very sensitive to ram speeds, it lost 10%! The reason for this is that it lumps all three bandwidth numbers and the latency together into one number, dropping ram speed without dropping timings makes the bandwidth worse and makes the latency worse. The result is a big drop in score.
In the real world you aren’t really likely to notice the difference unless you’re paying close attention and doing very ram related things. That said, if you can run 2000 I definitely would!
