Originally Posted by xenkw0n
It seems like LLC does more harm than good unless pushing serious overclocks because these chips get pretty warm well before pushing into the 1.4v+ range.
Settled my 4.4ghz overclock with LLC off and under stress it bounces between 1.264 and 1.28v. Without being fully taxed it hits up to 1.32v and overall there's just no need for LLC in this instance. If I wanted to push to 4.6 I could see it becoming useful but even at 4.5ghz and LLC off, 1.4v works as it droops down to 1.328 territory. I found myself using .02 additional turbo voltage to help combat the serious vDroop on this X79-Deluxe board.
Memory tightening is going to be a lot longer of a process than it normally would be because the system wasn't setting tertiary timings lax enough for me to just worry about what could be causing instability in the other timings. Was getting memtest errors (not a lot, but I don't want any...) with auto settings on my third timings and they went away after I loosened them up. Just means there's gonna be some small adjustments and then daily memtest runs when I'm at work until I hone in on what my sticks can do without risking data corruption. Bumping the voltages wasn't helping, they were way too tight... the memory training is impressive on this board but the memory is good enough to operate for initial boot with the aggressive timings. We're talking 1-2-3 and a few 0 timings set in the tertiary ones. None of them get set over 4 even on auto when setting to 2133 or 2400mhz.
Thanks for the advise on timings, i will try that.
Updating on my tests, I found some strange behaviours: the system boot and works normally, but it freeze sometimes with a kernel-power id:41.
I succeeded to replicate this freeze:
in Blender 2.79 I work on a model (around 1M vertices) with a window in 3d view (only surface mesh representation) and a window with render mode (so same object as above but with materials and lights, so gpu load).
From this situation if I open a molecular visualization sw (ucsf Chimera X) windows freeze. It does not happen if I have only the surface and not the render in Blender.
I tryed all of the prime95 tests for at least one hour, and no freeze happened.
So i'm wondering if the problem is due to a low voltage or some kind of cpu-gpu miscomunication due to the high bclk. (on oclmembench i found a lower bandwidth in oc)
ram: auto 1866Mhz, 1.656v (but hwinfo is showing me 1063x2 Mhz on loose timings).
Next step I will do will be to manually set the ram voltage and clock and try an higher vcore on the cpu.
I changed some setting:
ram multi: 16x (in windows it goes to 1824Mhz (912Mhz x2 as seen in HWinfo and in the BIOS....wmic memorychip get speed
command in the prompt reads 1600 Mhz)
seems stable, I will update with some more test soon.