It's him again! (everyone hide, lol)

[Aenra]

It's my nature.. just..

Everything was fine, but the OCD is strong with this one. So for no reason i could actually relate to you, i thought i'd try memtest pro again! I mean it's fun, right!?

Been at it since yesterday.. Errors, one or two every four passes, always in tests 7 and 8.
Of course, i knew what it was and indeed, once i changed Command Rate back to 2 it passed everything without a single error (all other timings and voltages the same). Straight-forward thus far right?
Except the thing is, Aida, XTU, etc they all gave me successful stress tests with T2 and said same timings and voltages.

I've read various posts where people say that if you pass those, you should ignore memtest. But i also remember Memtest's warning, it being that even with one single error, Murphy will get to you, sooner or later.
I am obviously opting for the safer route and leaving CR to 2.

Right?
(You can also be like, super nice to me and assure me there's no performance loss, the grass is greener with 2 and so on)

The autism society thanks you for your immense patience with this one.

 

[Solaris17]

To date that i know of there is no proper way to test memory inside an OS. 1 error can also lead to unforeseen consequences. i would just run it where it is stable.

there really is no justification. If the program says the hardware isn't running right then its not running right.

 

[cadaveca]

(You can also be like, super nice to me and assure me there's no performance loss, the grass is greener with 2 and so on)

You should be looking at resource monitor and checking for hard faults (if the number is high, usually it's related to unstable memory) as an indication as to whether memtest is wrong.... or not. You could measure performance by various ways, but they might not show issues such as hard faults, so simply looking at performance lost doesn't accurately describe what might really be going on.

You can also check Reliability monitor for issues.

 

[Aenra]

@Solaris17 i'm using the bootable UEFI version of memtest, in case this matters. I'm not doing it from Windows.

@cadaveca thanks, will be keeping an eye on it

 

[Regeneration]

Are you referring to MemTest86 or HCI MemTest?

MemTest86 isn't good to test stability during overclocking. MemTest86 meant to check the integrity of memory modules. It doesn't apply enough stress.

To test memory overclocking, use AIDA64 (latest version) or HCI MemTest (disable HT, disable Windows' pagefile and run 8 instances).

Anyhow, sometimes MemTest86 have "compatibility" issues with some hardware. If you insist on "testing" outside-of-the OS, you can try MemTest86+.

 

[Solaris17]

I'm not doing it from Windows

AIDA and XTU are in windows.

 

[Aenra]

@Regeneration good point, should have clarified in advance. Am using this one:
http://www.memtest86.com/technical.htm
Am using the Pro version, last updated Feb 2017.

As to the rest, i have Aida, XTU and Prime for inWindows testing. And in case i phrased it poorly in the OP, i pass them all. I made the thread because while they all tell me everything's fine in the Shire, memtest insists there are errors.
(i mentioned the non-Windows part in the following post because memtest isn't run from there. You make a bootable USB drive, load it from BIOS override)

 

[m0nt3]

I had an issue with our skylakes at work (optiplex 5040's), where memtest would freeze in the same spot. I had read there was an issue with memtest and the skylake platform, cant find where I read that at the moment. Will post if I find.

 

[Aenra]

Not sure if it would affect me in any case, different chip. Would be nice if you did though

With what i've got so far, what i'm thinking is that the safest route for me is to stick to T2 (to be fair, i've been using memtest before some of these other tests even existed, am willing to trust it), and take advantage of the extra voltage it required so as to make my timings even tighter; hopefully even 13-14-13.
Whenever i get the time again anyway, these tweaks take too damn long, lol

 

[Regeneration]

Again, try using MemTest86+, to ensure it's not a bug in MemTest86, but take in mind that both MemTest86 and MemTest86+ are meant to scan memory modules for defects (NOT STRESS TEST). To check for overclocking stability, disable Windows' pagefile and Hyper-Threading, and try HCI MemTest and AIDA (at least 6 hours each).

 

[Aenra]

Dude i'm thankful for your replies, but i honestly doubt you've read through my posts, i'd only be repeating myself.

 

[basco]

test 5 was\should be the hardest so normally if it gets past 5 you should be clear for test 7+8 but maybe they changed some routines.
and other Q: do ya really need 1,250volt for 3250 cache??

 

[Aenra]

@basco as i recall (may well be mistaken), test 8 is basically an expanded/more thorough test 5; it's meant to catch those that got away from 5. Been a very long time however and what posts i could find on the subject are also dated. Maybe things have changed?

As to cache, yes and no. With RAM at stock, needed a lot less. With RAM where it is, i had to boost it further, along with SA. It does look like too much, but that's where i can pass everything :S
(i was failing the RAM stress tests regardless of dram voltages and at some point i happened to read a post in overclock.net/f's DDR4 overclocking thread, advising bumping the vring when tightening TRFC. Did the trick for me. Default TRFC was 600, lol)

 

[JunkBear]

Why not using thé memtest onto the Windows7 repair dvd? Even if you use 8, 8.1 or 10 it still starts onto the DVD and can test the memory.

 

[Aenra]

@JunkBear you mind rephrasing that? I do not follow, sorry :S

 

[JunkBear]

Create à Windows 7 repair disk using à Windows 7 machine. Then you boot thé computer with that repair dvd and choose the function to test the memory.

 

[Aenra]

Not sure why you'd advise that, especially the Win7 part, but either way? Not happening. I'm not gonna switch my entire OS when i have an abundance of tools already available to me

(thank you for the advise however)

 

[EarthDog]

Sorry... why did we take the red pill here?

You pass everything except some random memory test and don't seem to have any issues in anything you do so........?

Make thread titles which describes the problem instead of being 'cute'. This helps tpu and others searching for the same issue. As it stands, nobody can get any answers running across the title.

 

[Aenra]

I am only asking 'cause i don't know basically..
(and as such i'm not sure if i should be worried or not i mean)

 

[EarthDog]

Move on Aenra. Don't get lost in the minutia. Seems like you are ok so far.

 

[Aenra]

Yessir

 

[qubit]

The fact that memtest fails one of the tests and everything else works does suggest that the PC isn't 100% stable at that timing. I would therefore use the faster timing, but if the PC behaves weirdly or crashes, always suspect it and set it back to troubleshoot. Heck, it might help to swap the modules round, you never know.

I've had overclocked PCs which wouldn't run the most stringent stress tests properly, but everything else was fine, including all my AAA games, so I didn't worry about it. It mostly matters if you're running things like Folding@Home, Prime95 and other CPU intensive stuff.

 

[RejZoR]

To date that i know of there is no proper way to test memory inside an OS. 1 error can also lead to unforeseen consequences. i would just run it where it is stable.

there really is no justification. If the program says the hardware isn't running right then its not running right.

Yes there is. Hammer your system with ASUS RealBench Heavy Multitasking test which taxes memory a lot. And 7-zip benchmark with nearly all memory allocated. Use huge dictionaries. This has proven most reliable when I was tweaking my RAM.

 

[Aenra]

@RejZoR you've misunderstood. I believe he's referring to Windows needing a portion of your RAM to function plus a secondary one in reserve. Hence his saying "proper" + "inside an OS", ie that one cannot possibly test it in its entirety from there.

(the extent to which this matters of course is another matter. Not one to judge on that)

Anyway, am taking EarthDog's+qubit's advice.. i think i've spent way too much time on this as it is, lol

 

[RejZoR]

You think Windows occupying section of RAM isn't stressful by itself? Because when things go bad, Windows will be the first one to tell you. Especially in case of ASUS RealBench, I wasn't getting hard crashes or BSOD's, it was ASUS RealBench that started saying that files became corrupt. Meaning it has its internal integrity checking and when things go weird, it'll tell you. If you get BSOD or hard lock/reset, it's self explanatory. ASUS RealBench is my first go to stability checking. For CPU I just let it make 10 loops of H.264 video test. If it passes, it'll be stable for sure. For memory I use Heavy Multi-tasking which plays video all the time and then does cycle of video encoding, data compression and image compression. And repeats this for 10 loops. No need to make whole night stability tests, this tells you in like 10 minutes. It's by far the best tool imo and I've used it for all my CPU/RAM overclocking testing.