DDR3-1600 sticks unstable? at XMP settings

j5v

Honorable
Feb 14, 2014
12
0
10,520
I'm running: ASRock Z87 Extreme 4, Intel i5 4670K, and 2 x 8GB DDR3-1600 9-9-9-24 1.50V.

It is very unstable at the main board's own "Turbo 4.2 GHz" profile (paging errors, and consequential filesystem errors) but no faults are found by memory tests. Manually changing the DDR3 to run at 1333 makes it much more stable, with two blue screens in a week.

Blue screens usually occur when the system is not under load.

What should I tweak first to troubleshoot this?
1. Relax the RAM timings a bit, 1600 10-10-10-27?
2. Up the memory controller voltage,
3. Reduce the memory controller's speed
4. Remove the overclock entirely, and try XMP presets again?
5. Something else?

At what point should I think the RAM might not be up to its rated speed?

p.s. The "Turbo 4.2 GHz" setting gives:
CPU All-core ratio: 42 x 100,
CPU cache ratio: 39
I locked BCLK ratio to 1, and BCLK/PCIE to 100.
CPU Input Voltage is 1.9V, LLC level 2.
The adaptive voltages are set in such a way that editing cannot reproduce the values.
 

snowctrl

Distinguished
Do u mean yo Vcore is 1.9? This is WAY too high... set it t manual rather than auto in BIOS, then set it at 1.3v - u don't ever want it any higher, or u will shorten t life of t chip.

At 1.9v, this will likely b t source of all yo issues. Yo PC should NEVER blue screen.
 

j5v

Honorable
Feb 14, 2014
12
0
10,520


*grin* No, I wouldn't do that. Definitely "CPU Input Voltage" 1.9V.
VCore is a safe 0.7V idle, maxing at about 1.13V-ish under ordinary load 4.2GHz, and 1.23V under burn-in load (extended instructions).
 

j5v

Honorable
Feb 14, 2014
12
0
10,520
Solved!
XMP settings failed.
Instead I went with JDEC setting for 800Mhz (1600), which worked.

DDR3 Binning

My main board has a 900MHz (1800) setting, so I set the other timings to be approximately the same in absolute terms, with a faster base clock: 1800 CL10 (10-10-10-27).
The binned speeds for "Patriot Viper 3 Black Mamba" are 1600, 1866, 2133, 2400), so there's a chance that 1800 would work for my 1600-binned item, and indeed it was stable at 1800. 1866 failed, as expected. There might be some tweaking room here, but the benefit would be negligible.

Further note on stability

I use a Scythe Mugen 4 (slow fan edition), with the motherboard throttling CPU clock by its power usage, which keeps it well below TJMax, and never invokes the CPU's own throttling. This gives it a higher max multiplier (43x at ~60°C for most actual workloads, e.g. POV-Ray), and lower multipliers for burn-in tests (IBT: 38x-43x = 107 GFLOP). This seems a bit more fit-for-purpose than what I've seen in general overclocking advice.