Can Memtest86 be wrong with the new Z68 chipset?

[keplenk]

Hi all,

Greets! I recently built a new sandybridge PC for a friend and here are the specs:

Intel Core i5 2500K
GIGABYTE GA-Z68MA-D2H-B3 http://www.newegg.com/Product/Produc...-495-_-Product
Memory: GSkill Sniper 8gb http://www.newegg.com/Product/Produc...k=sniper%208gb
Corsair CX600
EVGA GTX460 1gb SC

After installing windows 7 (which installed fine), I ran memtest86 just to see if there are no errors with the RAM. I used a linux mint live cd which comes with a memory test and to my surprise, the memtest that came with that distro doesn't even run - it crashes right away. So I went and downloaded the latest versions of memtest from memtest86.com (versions 4.0 date released 28-March-2011) and memtest.org (memtest+ version 4.2 25-January-2011) burned the ISOs to disk.

The first thing I noticed is that memtest+ 4.2 doesn't detect the RAM settings correctly - It thinks that I have Triple Channel and CAS of 6-6-6-20 - but its supposed to be DUAL channel and CAS 9-10-9-28. On the other hand, memtest (not with +) detects my i5 as an i7 and has "unknown" caches.

Unfortunately, both memtest and memtest+ gave errors on my Sniper modules. So I ran one stick at a time to see which of the two dimms are causing the problem (memtest86+ detectes as Single Channel but still CAS 6-6-6-20 instead of 9-10-9-28)

Things are so weird with my results and for better understanding I will name my two x 4gb Sniper modules A and B.

Here are the results:

XMP enabled running at 1866mhz
memtest 4.0 = both DIMMS gave errors (after 2 passes)
memtest+ 4.2 = DIMM A had no errors // DIMM B had multiple errors

XMP Disabled running at 1600mhz
memtest 4.0 = DIMM B had no errors // DIMM A had some errors -- (WTH! the exact opposite of the XMP enabled 1866 memtest+ test)
memtest+ 4.2 = both does NOT have any errors (2 passes) --> this is the only time no errors were reported in all my tests.

I tried different DIMM slots just to be sure (all slots gives errors - 2 passes each slot).

As you can tell, all tests have very mixed results and I'm starting to believe that memtest/memtest+ could not just utilize the Z68 chipset yet.

The next thing I did made it a lot weirder. I used my own PC and swap parts with my friends PC. My motherboard is a MSI P55-GD86 and I use G.Skill Ripjaws 4 x 2gb DIMMS 1600Mhz. At first, I tried using the Sniper on my P55 board and memtest+ detected DUAL Channel and CAS properly. With my P55 board, there were no errors at all - XMP enabled/disabled. On the other hand, memtest (not with the +) failed to even run on my P55 board - I wonder why.

I placed my Ripjaws to the Z68 board and ran with XMP enabled and disabled. Both didn't give any errors.

Just to make sure it is not the Power Supply, I swapped MY Corsair TX650W with his CX600 - its not a power supply issue.

I didn't do any voltage changes on the DIMMS or anything as I haven't done anything like that before. I just know that the Sniper is rated and running on 1.5V on the Z68.

BTW, I'm not having crashes on Windows 7 and CPU-Z detectes CAS properly inside windows. Everything is actually fast inside windows.

Could it be the motherboard that is faulty? Memtest/Memtest+ error? The Dimms? Just needs a voltage bump? or nothing is faulty?

Is there any other memtest-like test that I can use just to be sure that the DIMMS are good? Within Windows 7 would be nice.

Please help!

Thanks.

PS: If I need a voltage increase, could someone please teach me how to do it? What voltage increase is safe? I'm currently running at 1.5 right now.

 

[Elixer]

Yes, Memtest could be reading things wrong, since that chipset just came out...
Windows has a memory test available as well, you might want to try that one.

http://www.sevenforums.com/tutorials/715-memory-diagnostics-tool.html

 

[keplenk]

Thanks for link

 

[lopri]

Memtest 4.20 worked for me with a 2500K and a P67 board. Make sure you are not overclocking when testing memory.

 

[keplenk]

Yes, Memtest could be reading things wrong, since that chipset just came out...
Windows has a memory test available as well, you might want to try that one.

http://www.sevenforums.com/tutorials/715-memory-diagnostics-tool.html

Windows' own memory diagnostics also gave an error

However, I just found something interesting. When I plug a single memory module (only one DIMM) to the motherboard, memtest detects it as 6-6-6-20 no matter what. If I add another memory module (dual channel), that is when timings are reported properly (9-10-9-28) -- using the G.Skill Sniper 8gb 1866 RAM.

I even tried this with my own G.Skill Ripjaws 1600Mhz ram and displays the same issue. Plugging one only detects it as 6-6-6-20 instead of 9-9-9-24 (again, this is for ripjaws). Adding another module fixes the timing display inside memtest+.

On the other hand, even though memtest+ thinks that running a single DIMM has a timing of 6-6-6-20, when I go to the BIOS and Windows (CPU-Z), the timings are reported correctly.

I'm now not sure if those timings are just cosmetic errors of memtest+. If its not, it might be the reason why the Sniper 8GB throws out some errors. As a matter of fact, errors given by memtest/memtest+ are very random. Sometimes, it throws over a 100 errors before a single pass. Sometimes, it throws 4-5 errors. Another update is now is that both the Sniper 2 x 4GB (run one at a time) give errors. Before, one Sniper 4gb was clean. For some reason, it throws errors now.

Again, the G.Skill Sniper when tested in my own MSI P55 motherboard, didn't throw any error at all (ran it for more than 8 hours) with memtest. On the other hand, my Ripjaws 1600Mhz was tested on the Z68 board and it didn't give any error also (ran for 8 hours with 14 passes).

On a side note, despite memtest+/memtest throws out hundreds of errors with the Sniper 8Gb memory. I've reinstalled windows, installed a bunch of applications, I even copied a single 8GB iso file back and forth to my external hdd (ran md5 checksum afterwards) and there were no errors. No BSODs, No hiccups. The system is running perfectly fine.

Since im still on the 30-day window return period, I'm not sure if I should RMA the memory or the motherboard.

As I can see from the weirdness I experience it could either be the two.

What do you think guys?

EDIT: UPDATE: I just confirmed that the memtest+ 4.2 reporting 6-6-6-20 ALSO occurs with the 2 memory modules installed BUT it is running on SINGLE Channel mode. In dual channel mode, the correct timings are reported. I've tried both with Sniper and Ripjaw.

 

[Elixer]

You can just be getting lucky, and not be running into the troubled RAM area yet.

Anytime I get any error on memory tests, I RMA those modules.

 

[keplenk]

You can just be getting lucky, and not be running into the troubled RAM area yet.

Anytime I get any error on memory tests, I RMA those modules.

Thanks for the reply. Exactly my thoughts. I don't want to risk it and regret it in the future. I might RMA RAMs for now and I'll see from there. I'll avoid 1866 speeds for now.

Thanks!

 

[sticks435]

It might also be the motherboard. My current P35 DFI board went through 3 sets of Crucial Ballistix before I switched to Corsair and even those failed when testing them with more than 1 module. Certain slots threw less errors, but I finally gave up. In hindsight I should have RMA'ed my motherboard, since I can only get a 3.2 OC on a Q9550 without errors.