Early Memtest86+ Test #2 Errors - Then Vanished

[guptasa1]

Hey all,

Weird problem - well, hopefully not a problem, but curious nonetheless...

Was having problems getting stable memory speeds on my motherboard, so tried a few different kits/timings and making progress. Finally found what I think to be stable memory (passed 30 complete passes Memtest86+ and is close to 24 hours on Prime95).

However, I did get about five (all at once) Memtest Test #2 errors with the new RAM configuration, but only first run, and only briefly, and can't reproduce it now.

I went into the BIOS immediately afterwards to make sure everything was right (it was), then retried the test with the same exact settings. It was fine. Just to be on the safe side, after retesting briefly, I went ahead and reseated the modules anyways and also set voltage to Auto (as I had it set to 1.9, which is max supported, but probably don't need that much). This time it went 30 complete passes (15 hours or so) with no errors whatsover. I also started it again from a cold boot a couple times to make sure it wasn't a first run thing.

Should I be concerned about the errors I saw as long as they don't reoccur? Any ideas what might have caused them? I think I was putting the case back on and doing a bit of cable organizing (not unplugging; just trying to move them slightly for airflow) at the time but shouldn't have affected anything. Conceivable I touched the mouse/keyboard too (as it's on the floor next to the system right now).

Thanks!

 

[TC91]

rerun memtest for an extended period of time (overnight or 24 hours+) to check if they reappear, otherwise you should test prime 95 blend for each cpu core in windows for a few hours to see if anything goes wrong.

 

[MarcVenice]

Perhaps the timings weren't right, or the voltage wasn't high enough. I'd dare say if you didn't get any errors on the second run of memtest, and none with prime95, you're good to go.

 

[guptasa1]

Well, so far...

- 30 passes of Memtest86+ (15+ hours) with No Errors
- 26 1/2 hours of Prime95 Blend on 4 Cores with No Errors
- 3 Hours Memtest86+ Bit Fade Test (1 Pass) with No Errors

Still to go:
- 1 Hour OCCT
- 1 Hour Furmark Xtreme Burning
- SuperPi 32 Million Digit Calculation

I really feel this thing's probably stable... This is *definitely* the best I've done so far and the only kit that's liked me (at rated speed; I refuse to run a kit at less). I know the values were right in the BIOS, which is what confuses me, but maybe it was just a strange boot or they didn't take the first time or something. So strange.

 

[QuixoticOne]

It is more stable that most would probably accept.

Personally I'd be a little concerned, though, and I'd run truly extended testing on it during any time
the computer is not needed (e.g. overnight, during workdays, whatever), and I'd do that for around a week
or two which is nice because then you get to burn everything in and further reduce the risk of infant mortality
going undetected during your return / warranty period.

I'd also suggest OCing or tweaking until you're at a point where you feel you'd be happy for long term
operation, then document all that, leave it there, and do the long term tests.

It could well be just something like a BIOS bug causing mis-set timings / voltages, or an AC power fluctuation / surge,
or ESD due to your environment, or so on. But I'd test to verifiy that it really seems stable over days/weeks just in case.

It is possible you have a truly intermittent problem like a bad solder joint that intermittently makes contact in the
board in which case it could be fine for months then break with no warning.

It could be that the socket wasn't making good contact and reseating it or allowing it to settle cured it.

It could also be cosmic rays or such unpreventable glitches; it is expected to see a few glitches every few
weeks due to those kinds of things; that's why they have ECC Registered RAM for servers.

If you have auto updates on or have manually updated the BIOS, it could be that an update helped too.

 

[QuixoticOne]

OBTW if you really want to stress it, run the system hot, somewhere around 40C case temperature,
that should make the likelihood of some kinds of failures increase and better show any marginal operating
stability.

A cold test can help further confirm operations but usually that's hard to do without an air conditioner and small closed room.

I've got one known bad DIMM that only fails memtest about 20% of the time, so it is possible to have it
succeed for several runs then fail.

 

[guptasa1]

Thanks for the great feedback - good to hear I'm probably okay.

The computer isn't actually in use yet - I've been struggling for weeks to get 1600 MHz memory stable (the annoying thing is...I'd get errors but only after 8 or 10 hours in Prime95 or a bit more often in Memtest), so it's pretty much been *constantly* running some sort of test for the past 3 weeks straight (ordered the parts the week of the 4th and built it the following weekend). But aside from Memtest Test 4 and Prime95 errors I was getting from the unstable memory (or too aggressive timings), I haven't had any problems. No real world crashes in Windows, blue screens, or anything like that.

Really considering finishing the tests tonight and starting to use it tomorrow or Wednesday. That should give me a real-world perspective. Plan to game a bit too which should give me a different sort of stress test (though I already tried a few demos).

 

[guptasa1]

Just as an update, I was successful on the following (no errors):

- 30 passes of Memtest86+ (15+ hours)
- 2 passes (6 hours) Memtest86+ Bit Fade Test
- 2 passes (several hours) Windows Vista Memory Tester Extended Test
- 26 1/2 hours of Prime95 Blend on 4 Cores
- 1 Hour OCCT Auto Test
- 4 Hour OCCT Custom Test
- SuperPi 32 Million Digit Calculation (crashed once, but tied that to a Vista issue after doing a search as the program actually stopped working instead of making a math error - ran it multiple times after I got it working to verify it could do it)
- 1 Hour Furmark Xtreme Burning Stress Test
- 1 Hour Furmark Fullscreen Normal Test
- Passed Windows Experience Index @ 5.9 Everything

I think it's pretty safe to say at this point this thing's stable. Hell, it even comes out of standby/sleep. Never had a computer that could successfully do that before. =oP

 

[Zap]

IMO how worried you should be depends on what you use that computer for. If a mission critical system for a huge corporation, then be worried. If just gaming and net browsing, then you're fine.

I had a similar experience with my LAN party box. Suddenly Windows wouldn't load and I couldn't even reinstall it. Memtest failed hard, and I had to underclock my memory to even get Windows back on there. Then, couple weeks later I clocked it back up to stock and it passes Memtest.