Four memory modules = HDD data transfer corruption

[The Keeper]

I've got four memory modules of Kingston 2GB DDR2 ValueRAM in Gigabyte GA-P35-DS3. Memtest86 passes fine, no errors. So does Windows 7 memory diagnostics on standard settings. But when ran with extended settings, it does find problems but doesn't specify any better than that.

In actual use data transfers between HDD's or even file copies within same HDD gets corrupted if I have all four RAM sticks installed, this tends to happen only when transferring large files such as closing to 500MB and larger. Bigger the file, the more certain data corruption is. This is only few bits and bytes here and there, but enough for file hashs to notice, particularly 7-Zip when extracting files. Very rarely does other applications notice any issues though. I've gotten data transfer issues from DVD to HDD as well, such as when installing games. The DVD drive and all HDD's are SATA.

Still, data corruption does happen with four sticks. No issues with just two, even Win7 extended memory diagnostics passes with just two sticks installed. Tested all four memory slots as well. Is there anything I could try to fix the problem aside of getting new hardware? I do not have any overclocks and I even tried to increase RAM voltage by 0.1 to see if that'd actually help, but no such luck. I've also tried AHCI on and off without any difference. Any ideas people?

 

[VirtualLarry]

I wonder if that's the issue that I'm running into.

http://forums.anandtech.com/showthread.php?t=2174024

 

[RaiderJ]

Do any two stick work fine? Wonder if you've isolated it down to four sticks running, or if you could find a bad single stick. If it's just all four at once, try loosening some of the timings to see.

 

[The Keeper]

All sticks and memory banks have been tested two at a time and they work fine as long as only two are used. Timings are set to be automatically detect and use what SPD reports in the BIOS and I double-checked that they're also what Kingston's datasheet says.

Guess I could try lower memory timings though.

 

[Phynaz]

Is everything at stock?

 

[The Keeper]

As mentioned in the original post, no overclocking.

 

[Phynaz]

Man, I'd have to point at a bad motherboard.

 

[Emulex]

no kidding. but don't dismiss bad ram! this is where registered and ecc ram come in handy the bus load with 4 dimms of questional quality "unbuffered" is tremendous.


what happens with 1,2,3 dimms ? is it cool?

 

[The Keeper]

There shouldn't be bad RAM because all four sticks have been tested two at a time, as have been memory slots two at a time. Haven't tested three sticks/slots because memory would work in single-channel mode then.

I will probably try to get used P45 motherboard to replace this Gigabyte P35. Right now looking at auction of Asus P5Q, I'll probably end up having to pay about 70-90€ for it though which is a lot for older used mobo.

 

[bryanl]

More modules mean more capacitive loading of the memory bus and worse signal integrity. Some BIOSes automatically compensate for this, by switching to slower timings (a BIOS upgrade may address this better), but some motherboards are simply speced for slower memory when more than a certain number of modules are installed and require manually going into the setup and selecting slowing timings.

Kingston is one of the many manufacturers known for pushing the envelope when it comes to meeting specifications, often approving parts that Samsung or Micron would reject for lacking sufficient quality margin. IOW test your system with 4 modules made with Samsung or Micron chips.

 

[The Keeper]

BIOS is latest available. I probably should try slower timings then, but I do not have any other DDR2 sticks to test with. I've noticed that timings are no different whether I have two or four sticks installed though.

 

[BTA]

I had an issue with my wife's Gigabyte AM3 board and SPD timings of some Kingston memory. For some reason it will not set to SPD, it actually sets the timings lower than they should be. I had to manually enter the SPD settings. It's pretty retarded since it actually shows the SPD in one column and then the wrong ones it set next to those. Works ok now.

So like the others said, I would try hard setting the timings or even bumping them up a bit.

 

[The Keeper]

Right, so default timings of these Kingston sticks are 5-5-5-15. I tried to increase these to 6-6-6-18, but the problem persisted.

Next I tried what happens with three sticks installed. I tried stick #3 on 2nd and 4th bank, no problems. Then I tried stick #4 on 2nd bank, data corruption. Then I tried it on 4th bank, again data corruption. Puzzled by this finding, I decided to try stick #4 together with one of the other sticks. No data corruption.

What the hell? I really don't know what conclusion I should draw from this. Is it the RAM stick after all?

 

[max347]

How long have you run memtest? Though it would take alot of time, this would save you money-
Stick one in slot one for 8+ hours
if pass, stick one in slot 2 for 8+ hours
if pass, stick one in slot 3 for 8+ hours
if pass, stick one in slot 4 for 8+ hours
then you will know if any slot is acting up (never seen a problem in a desktop, but have in a laptop).

Then, test each stick for 8+ hours alone.

I know you have already done it, but I have memtested systems where 2 passes ran fine, but running it overnight revealed errors.

So, if all pass these tests, you know that your sticks are good, and your board's single channel is working. Next I would take 2 sticks, and do 8+ hours in both pairs of slots (8 hours in slots A1 and A2, and then 8 hours in slots B1 and B2).

Now, this is obviously alot of testing, but really the only way to definitively rule things out. What kind of PSU do you have? Have you opened up HWMonitor and checked for voltage fluctuation?

Any chance you have access to spinrite? For any harddrive related issues I always run a scan with it (I am sure there are other utilities you could use, but I have found spinrite to be my favorite).

If you are not overclocking anything, to avoid the hours of memtest, I might run LinX with your memory setting at "all", and see what happens. That is one in-windows utility that has found memory errors in systems that seemed perfectly stable.


/speculation

 

[The Keeper]

I have ran memtest86 succesfully overnight with all four memory sticks installed. Only Win7's own diagnostic program was able to find issues, but even that only with extended settings and only with four sticks installed. Haven't ran it with three sticks installed though.

I've found that copying 20GB file from HDD1 to HDD2 and then checking SHA-512 checksum is much faster test, since memtest86 doesn't find anything anyway.

Haven't checked for voltage fluctuations. Don't have spinrite. I have 80plus 500W Antec EarthWatts. Might give LinX a go later, thanks for the tip.

 

[max347]

Wow, very surprising. I was betting memtest just hadn't run long enough, but if 4 at the same time pass overnight, I would look elsewhere. I doubt an Antec would cause any issues, but I guess you can check it out.

I am at a loss here, other than a mobo issue. I always reserve that though, because it seems like such a catch-all. Unfortunately I have never had your exact problem, so I can't be of much more assistance.

Good luck!

 

[The Keeper]

Yeah, the problem is weird since it doesn't clearly point to any particular hardware. I might give up on having 8GB RAM in this rig, 4GB is still quite enough. Well, teaches me to stick with two memory sticks in the future as well.

Heh, maybe I should start planning on Bulldozer or Ivy Bridge rig...

 

[bryanl]

I have ran memtest86 succesfully overnight with all four memory sticks installed. Only Win7's own diagnostic program was able to find issues, but even that only with extended settings and only with four sticks installed. Haven't ran it with three sticks installed though.

The normall correct practice is to declare the memory faulty if it fails any test at all, provided the other hardware is known to be in good working order and compatible with that memory.

Slowing the memory clock or increasing the command time can eliminate errors when nothing else works, and the command time will have almost no effect on performance. Not all BIOSes allow the latter to be adjusted, so some people have resorted to the risky practice of reprogramming the SPD chip of each module.