While it's detecting the full 10GB at 'loose settings', I'd do a
memtest86+ run. 8 full passes should be sufficient enough to uncover any apparent dud dimms/modules.
This should help differentiate between potential hardware faults vs settings issues.
(Let it run overnight as 8 passes will take around 8hrs - although if you get errors quickly, halt the test. Then you have the 'pleasure' of removing/swapping modules to test dimm/moudules slots.)
If there are zero memtest errors and it still consistently maintains the full 10GB detection on a cold boot with the current settings, you can slowly start changing some of the settings back. However leave the DRAM volts at 1.64v as you'll need this to run @ 1600mhz
►Start by upping the mhz back to 1600mhz. *NB the UCLK frequency should be double what the DRAM frequency is. AUTO
should do that automatically, but it's probably worthwhile setting it manually.
► Set the QPI link up one notch.
► *Optional* Set the CR back to 1N (although there really isn't that much noticeable difference in terms of speed between 1N and 2N, and with a full 10/12GB 2N is better for stability )
►After that you can start lowering the QPI/DRAM volts until it either fails to detect the full amount, or becomes unstable. Although 1.35v is perfectly fine.
If after adjusting any of these settings results in detection loss, go back to a lower setting.
(Depending on the board/RAM - 6GB kits are '
guaranteed' to run at the listed specs ie 1600mhz @ 9-9-9-24 1N, but are '
not guaranteed' to run at those timings in a 12GB config, hence the need to loosen some timings / increasing voltages, particularly the QPI/DRAM voltage etc)
However, if there are zero memtest errors with the full 10GB detected (basically eliminating faulty hardware), but it still occasionally fails to detect the full amount - I'd start looking at other culprits like the CPU cooler tension which affects the CPU pads / socket pin contact etc.