Errors in WINDIAG memory diagnostic

[Team42]

Hi folks

For the first time ever, I have run a memory test using WINDIAG (yeah, yeah, I'm new to this overclocking lark. Gimme a break please...).

My memory is 4 x 1 Gb of Corsair CM2X1024-6400C5DHX running at 4-4-4-15 Vdimm 1.968v (stock spec is 5-5-5-18 Vdimm 2.1v).

At the first attempt, the diagnostic showed a number of errors in each of the 6 standard tests. On inspection, these were confined to Bank 6/7 (slot A3) of my Gigabyte GA-P35C-DS3R mobo. I re-seated the apparently offending module and tried again. Same result. I swapped modules around. Same result. Then I took the modules out of slots A1 and A3 and tried again, hoping to find 2 x 1 Gb of RAM in dual channel mode error-free. (The mobo spec allows dual channel mode if slots A0 & A2 or A1 & A3 or all four slots are occupied).

But oh no. Now I have errors in Bank 4/5 (slot A2).

My confusion is thus: the module originally in Bank 6/7 is no longer in the system and Bank 6/7 is no longer populated. Bank 4/5 did not previously show any errors. How come I have errors where none previously existed?

I have only just started to swap things around, and will be running more tests, but I'd be grateful for any advice on what could be the possible cause/cure.

Should I:

(a) start again using stock specs (timing and voltage) (the system had passed 10 hours of Orthos without error)
(b) re-populate all but one slot and run the diagnostic again, then change the unpopulated slot and retest, and...well, you get the picture...
(c) [insert cunning ploy to make my system work beautifully - please provide detail]

Please form a disorderly queue with all the right answers.

Sheers

T42

 

[Mondoman]

First, I would start with a dedicated memory testing program that doesn't require Windows. memtest86+ is freely available as a bootable floppy or CD image. Once you can run it through at least a couple of complete passes without any errors, move on to a stress test in WIndows such as Orthos (running in torture-test "blend" mode). If you can run Orthos for an hour or so without errors (or overheating - watch your system and CPU temps), you've got a pretty stable system.
Second, as you suggest in option (a), your errors likely come from running the RAM at a lower voltage (and faster timings) than the module manufacturer has certified. Note that these are technically "manufacturer-sanctioned overclocking specs", not "stock specs", as DDR2 stock spec is 1.8V.

After you get your system configured properly, I'd strongly suggest reinstalling Windows and your other programs, as the errors you've recently detected mean there's a good chance your files have become corrupted.

 

[Team42]

Cheers, Mondoman.

I've reset timings and voltage to factory spec (5-5-5-18 Vcore 2.1v) and run Memtest86+ v1.70. And I've got lots of errors already. Looks like tomorrow is going to be a busy day for me trying to isolate the problem(s). Failing address is 0007feedf18.

When I start getting errors, is there any point in continuing the test? Should I stop it and try reseating/swapping the DIMMs before re-testing, or is it more beneficial to allow the test to run longer?

BTW, WINDIAG doesn't run in Windows. It operates from a bootable CD, just like Memtest as far as I can tell. It loads and runs before the PC boots into Windows.

Ta.

T42

 

[Mondoman]

Thanks for the Windiag info. Anyway, if the problem doesn't show up with any 2 DIMMs installed, but only with 4 installed, then you might need to raise the northbridge/MCH voltage a notch. Consumer systems use "unbuffered" DIMMs, which means that every additional DIMM installed results in the control signals to each DIMM being weakened, as they are split among more modules. It's somewhat common to have issues when using 4 DIMMs.
As for the testing, go ahead and stop when you see an error -- there's no real benefit in looking for more, as one is already too many.

 

[Team42]

Ok, a brief update: I tried Memtest 86+ v1.70 and same issue. Lots of errors. I contacted Corsair support who advised me to test again at stock settings, which I did and got the same issue. Getting a little cheesed off, and out of desperation, I googled the error and...

I'm running Memtest again. Whereas I used to get loads of errors within 55 seconds of the programme running, I've got past 10 mins so far without a single error. Happy days!

The solution? I reverted to BIOS v F8. It seems that other people using Gigabyte P35 boards are discovering there's a bit of an issue with F9 (which I was using).

I love the internet. Is there anything it can't do?

 

[Team42]

8 hours of Memtest - 11 passes - no errors - big smiles in the T42 household.

 

[cozumel]

The F9 errors were not actually errors but were false positives incorrectly identified as errors when nothing was wrong. All the Gigabyte P35 range have the same problem with F9 BIOS. An update to the F10 BIOS similarly should remove the 'problem' as did your reverting to F8.

 

[Mondoman]

For those with slightly different MB models, note that my GA-P35-DS3P V2.0 was giving errors with 2x1GB Corsair XMS2 DDR2-800 RAM under memtest86+ with its F8 BIOS, but not with its F6 BIOS or F9 BIOS.

 

[Fadardo]

Also you should be aware that those windows memtests aren't 100% reliable.

Read this thread from Australia's top tech forum:

http://whirlpool.net.au/forum-...rchive.cfm/381565.html

 

[Team42]

Originally posted by: cozumel
The F9 errors were not actually errors but were false positives incorrectly identified as errors when nothing was wrong. All the Gigabyte P35 range have the same problem with F9 BIOS. An update to the F10 BIOS similarly should remove the 'problem' as did your reverting to F8.

Cheers Cozumel. I had almost concluded that (false positives) given that the computer actually seemed to be running fine (well, to the extent that it appeared to be doing what I asked it to do!).

T42