Memory - Instability/Crashes

rws

Distinguished
Dec 5, 2010
2
0
18,510
I've been having a memory issue but rather skeptical that it's actually the memory and not the processor or motherboard at fault. Memtest86+ reports errors on occasion but not very often, and windows memory dumps report a 1-bit corruption. I have 2x2GB GSkill DDR3 PC12800 modules installed. The problem occurs with 1 or both sticks installed and can occur in any slot. The strange thing is that I can move the slots the modules are installed in around until the problem stops. Then the problem won't occur again until I either reformat the PC or power it off. If either of those happen then I have to move the slots around again until I find 2 slots where the problem doesn't occur. It wasn't a big deal for awhile but the problem seems to be getting worse now.

Is this a motherboard or processor failure or did both modules start having faults at the same time (unlikely if you ask me)?
 
Your problem could be caused by one or more bad memory sticks or bad memory slots, a mobo electrical issue, a BIOS settings issue, or a psu problem. The way to diagnose it is to go back to basics.

- Clear CMOS**
- While the power is off, put the memory into the slots it should be in.
- Boot directly into BIOS setup, load defaults, and save.
- Exit into Windows.
- Use CPUID's CPU-z to determine what settings the mobo is using for your memory. Compare them to your memory's specs. If they are different, go back into BIOS and manually set the specs to match your memory (eg, 9-9-9-24, 1.5V).
- Exit into Windows and check using CPU-z again.
- Run Prime95 for at least one hour, using CPUID's Hardware Monitor to watch core temps. Do not allow temps to exceed 75C for too long. If you have a heat problem, resolve it. If P95 crashes or stops with a rounding or hardware error report, go back into BIOS and try adding one click more memory voltage, or relax the timings a bit if they were faster than 9-9-9-24.
- Repeat P95 test. If it crashes/errors again, the most likely problem is bad memory.

Let us know what happems when you follow these steps.

You can download the utilities using links found here:

http://forums.tweaktown.com/gigabyte/30530-latest-overclocking-programs-system-info-benchmarking-stability-tools.html

**Clear CMOS: Pull plug from wall, remove motherboard battery, press case power button several times to discharge capacitors, wait 5 minutes, replace battery, plug into wall.
 
Welcome to Tom's Forums! :)

Could be a lot of things {incompatible RAM, using XMP in the BIOS, GA MOBO, 'other' non RAM issues}. Therefore, please list the following:
1. MOBO {including revision & BIOS version}
2. RAM - Exact Part-Number(s), I prefer a link e.g.
3. OS {e.g. Windows 7 Pro 64-bit}
4. HDDs {SSD, RAID, etc}

We'll start there.
 

rws

Distinguished
Dec 5, 2010
2
0
18,510
I've already checked temps before, tested with a known good PSU & verified RAM settings. I havn't tried resetting CMOS or running prime95 though so I'll try that tomorrow night.

Computer specs:
MOBO: M4A79XTD EVO BIOS ver. 2001 (not 100% on BIOS ver. I'll double check that tomorrow as well)
RAM: http://www.gskill.com/products.php?index=221
OS: 32-bit Win7 Pro (I know not 100% of RAM is being utilized but it's not a windows issue as the problem prevents POST on occasion as well or BIOS will freeze)
1 WD HDD (no RAID)
 
OK, but note we're not trying to test temps - Prime95 will fail due to memory problems in many cases where memtest runs fine. My temperature warnings were there in case you hadn't done this stuff before. Temps are not likely the cause of your problem.
 

Crashman

Polypheme
Former Staff
Increase your RAM voltage to 1.60V, as specified from the manufacturer.


And never believe anyone when they tell you that Core i5 RAM won't work right on AMD processors. I've done enough testing to prove that such advice is always wrong.
 
@Crashman - apparently the Engineers at G.SKILL are inept. Please 'educate' them, take a vacation from your 'testing' job @ Tom's they really need your help.

G.SKILL Published Info:
F3-12800CL8D-4GBRM -> http://gskill.com/products.php?index=221
M/B Chipset:
Intel P55

F3-12800CL9D-4GBRL -> http://gskill.com/products.php?index=222
M/B Chipset:
Intel P55
Intel X48
Intel X38
Intel P4x Series
AMD 790 Series
AMD 785 Series
AMD 890 Series
 

Crashman

Polypheme
Former Staff


I can't read their minds, I only know what my tests have shown. But, I have seen some of G.Skills competitors use "incompatibility" of the newer parts as an excuse to push outdated, higher-voltage memory. Specifically, when I checked there was still a bunch of voltage-tolerant DDR3-1066 that could run something like DDR3-1600 CAS 8 at 1.75V, and that competitor would really rather someone buy this stuff so they completely separated their AMD and Intel product lines.
 
^ Both of the G.SKILL sets are 'current' and the $ difference is negligible.

Same goes for Corsair -> http://www.corsair.com/products/phenomii/default.aspx

Beyond that, I contacted both Corsair and G.SKILL and spoke to their tier-2 folks BOTH told me the same that certain IC's, not the voltage and comparing Apples with Apples, are NOT compatible with some different chipsets. So either they're ALL EVIL conspirators or there is some truth there?

I 'get' that you 'test' RAM -- BTW love to see some 'good' tests/benches/articles @ Tom's -- hint -- then according to 'you':
1. There's no such a thing or need for 'Matched Sets'
2. There's no need for QVL or Testing

---> hence you're out of a job-Right?! Are your tests how well they juggle or work as Dominoes or what?? What's the point of your job?
 

Crashman

Polypheme
Former Staff

1.) Usually not
2.) How would you know the whether the chips worked or not if you didn't test them?

Hows this grab you: JEDEC has standards and if your memory doesn't meet them, it's defective. You can add overclocking on top of the standards, but it should still meet the standard without overclocking. Intel and AMD both design their memory controllers to support the standards: If you put a memory and a memory controller in the same system that meet the same JEDEC standard, it should always work together.

I use "P55" memory on "everything", including X58 and AM3 motherboards.

Most likely problem with his G.Skill memory is that it's not operating properly at default voltage. Perhaps he changed the clock to its "rated" overclock without increasing the voltage to its "rated" voltage.

So basically what I'm saying is when he pops it in it should work, and if it doesn't work right there's something wrong with his memory (could have wrong default values) or motherboard BIOS (again, could have wrong default values). As long as the system boots and the chips or motherboard aren't bad, he can fix the settings in BIOS.
 

Per your answers:
1. What in the heck does "Usually not" mean? This is a yes or no question.
2. I asked specifically [There's no need for QVL or Testing] - what you said COMPLETELY contradicts everything you previously stated. And essentially CONFIRMS G.SKILL Chipset qualification. This too is a yes or no question.

As for a 'me' I own/lease about a half million in servers doing mission critical and multi-state relational aggregation for everyone from institutional to IDX for Brokers {REO, IDX, County Parcel and State litigation data} and I/we constantly test our servers integrity. 'This' is a diversion.

So I assume you purchase single, non-set, sticks or leave your DIMM slots empty on the X58... Again, a contradiction otherwise.

No, I won't disagree with proper BIOS settings or a bad stick(s) with the OP, but I too cannot ignore G.SKILL's info. What the heck is the 'purpose' of the Configuration and/or QVL - per you it seem they're ALL CRAZY?! However, I do NOT share your cavalier approach with RAM. If one of my staff "pops it in it should work" -> RAM into one of my servers he'd be out of there if they weren't fully spec'd; I never trust them again.

I do not share your views, or thoughts about RAM. I'd love you to Answer the 2 questions clearly and without being vague if that's possible for you.
 

Crashman

Polypheme
Former Staff


I think it's clear. 1.) Some manufacturers had to test their overclocked RAM in pairs to assure that one module didn't produce an error when overclocked, but that's an overclocking issue and better RAM consistency has otherwise negated the need for testing in pairs. The one problem that remains is when companies sell TWO DIFFERENT modules under the same part number, look to the next SBM for such an "unusual" find. 2.) Testing is needed by motherboard manufacturers to assure they aren't selling a defective board, testing is needed by RAM manufacturers to assure they're not selling defective DIMMs.

Per your other question, I have a bunch of dual-channel kits. When identical modules are sold in both 2-packs and 3-packs, three sets of two is the same as two sets of three.