Hi all again.
Seems my memory stick died or/and CPU IMC degraded...
This night I made some tests with CPU at 4.66GHz, keeping uncore at 3300-3400 range with 1.33-1.34v max, with different bclk and memory freq. I noticed that something is bad when I tested 186x25 combo trying to determine which Vcore is required for stable 4.66 Core, I finished with 1.4375v on Vcore testing CPU in Prime95 Blend while it produced a "memory_management_error" blue screen in about 30min of test. I turned pc off and started later again, but from the "verified" 4.5GHz range. And Prime95 started to fail each time...
And not as before - when some of 1-2 working threads stopped after a significant time like several hours (if there were not enough VTT voltage/or vcore), now this happens after prime95 starts to test IMC/memory stress problem size FFT size - above 1024k, it usually happens in 20-30 mins, when 1120k / 1200k blocks starts in test pattern and all workers stopping in no time - not 1 or 2 of them. Looks like something happens with memory and data corrupts or it "losts" I loaded my "verified" prime95 29hours stable config and it also failed in the same time the same way. Linpack were stable for 1 hour showing no errors and correct results, but when I stopped it manually - BSOD with "memory management error".
I run Memtest with CPU freq default (200x17) with high vcore (1.4v) and memory at 2000 with minimum uncore at 3000 and QPI/DRAM at 1.35v:
Then I dropped the bclk to default 133 and multi to 25, set all mem timings to auto, decreased voltages at minimum level and set mem to 1333 with 2000 uncore:
much less amount of errors (x12 less), and only from 7 & 8 memtest test patterns, not 9
then I put CPU at 1.6GHz 12*133, with 800 mem and then 1066 mem (1.5v), all with minimum possible voltages and memtest haven't find any errors... even 1
I put memory up again to 1333 and 1.65v on it and run memtest again:
NO errors again.
I allowed test to pass several times, turned pc off. Get in BIOS and set the 17*200 1.3v, 2000 mem 1.65v, 3000 uncore 1.3v and run memtest again, waiting to see errors again:
nope! I feel myself an idiot. Like my PC play with me but not I play with my PC.
Had someone faced such problems? Can this be defective or dying memory stick? If I understand right and if memtest test memory in blocks according to errorous addresses it should be 3'rd stick (the latest), because it's all that above 8192Mb up to 12gb.
Can it be an IMC? glitch? or degradation? but I think in case of degradation it can't come back from dead state and handle mem and uncore again as nothing happend?
I think I should take some mem. Could somebody give an advice, what of modern mem kits are suitable and optimal for our old 1366 westmere's/gulftowns? Newer "overclocking grade" kits seems to have high mem freq like 2000+ but bad timings like 11-11-11. While we need something that are able to handle both lower freq at tighter timings and high enough freq with sweet timings like 9 (as my current kit for example, it may move from default 2000-9-10-9 to 1800-1866 8-9-8 and 1600-1666 7-8-7).
I have old mem kit (forgot about it) that were made for MAC upgrades and is ECC and rated at 1066 freq (but mem sticks marked as 1333), the mem kit is from OWC. 3 sticks 4gb each. But I don't think that mem with ECC will go even to 1600. There are Micron chips. Finding info for them tooks only to some SO-DIMM modules used in notebooks and they all are about 1600 with timings like 11 and rated for 1.35v/1.5v...
In case of CPU... to have enough room and flexibility in BCLK I looking for X5680-5690. But they become too expensive last 1-2 months. The ES (engineering sample) are much cheaper and have to be unlocked multi, right? But how good they are?