Use memtest86 pass times as a performance benchmark for memory speed tuning?

[spkay31]

I am familiar with using memtest86 as an indispensable benchmark tool for testing memory stability and reliability. But I am wondering if when manually tuning memory speed and latency-timings fine tuning if the execution times of memtest86 test passes is a good measurment spec for overall memory speed? I am testing a G.Skill F4-3600C19D-32GSXKB with an XMP 1.0 profile for 3600 19-20-20-40. I have run this successfully at 3200 16-18-18-36 and 3333 16-18-20-38 but at a voltage of 1.30V vs 1.35V which is the recommended voltage at 3600 cl19. I realize this is a bit time consuming to test the minor peformance differences at these various speeds / timings but right now at initial system setup and benchmarking I am willing to spend some time maximizing performance and trying to make small concessions for better reliability (here I am assuming lower RAM voltage implies better long term reliabilty).

So since these speed/timing settings along with the voltage settings may have very small changes to overall memory speed and reliability, am I right in assuming the execution time for full test pass of memtest86 could provide a good benchmark for performance of memory at current speed/timings set in BIOS?

TIA,
-Scott

[David (PassMark)]

It is a very slow way to get a speed benchmark. But is should work if the rest of environment is held constant. (CPU, MemTest86 settings, etc...)

[spkay31]

Thanks David. Yes I realize this is a stability and reliability test, but I was interested if the execution times for say DDR4 2133 14-14-14-32 vs DDR4 3333 16-18-18-40 would be accurately reflected in the test results. I.e. the time to complete the test correlates closely to the memory speed and latency settings.

Naturally for a quick test I can run a simple AIDA64 memory speed benchmark, but I still need to run memtest86 after basic tuning to confirm stability/reliability of the speed and timings.

Cheers,
-Scott

[David (PassMark)]

MemTest86 has a built in memory benchmark function. No need to use AIDA64.