UPDATE: Memory tests fine in single channel, but not dual channel

[nervegrind3r]

to begin, I am not overclocking at all. The system can boot in to windows, but crashes at some point (within a few hours, and at no particular event). Usually, I get random bsods, reboots, and shutdowns out of nowhere. I have swapped the cpu's, psu's, video cards, hard drives, and removed all add in cards except video. I finally realized the memory is the problem.

Prime95 torture tests that test memory (2nd and 3rd test) fail within an hour, but the first test completes successfully (the one that doesn't test ram much).

Im using two idential sticks of Kingston 512mb pc3200 (cas 3). I always run it in dual channel, and always had problems. To narrow down if both sticks or just one stick was bad, I did thorough testing with memtest from a bootable cd. I tested each stick in both dimm slots for 24 hours each in single channel mode. Just to make it clear, I tested stick 1 in slot A for 24 hours, and then in slot B for 24 hours (single channel mode in each slot). Then I did the same procedure for stick 2. All four memory tests show no errors.

Then, I populate slots A and B with both memory sticks, and run the same memory tests. For whatever reason, I get errors when testing in dual channel mode. Im currently running my second 24 hour test in dual channel mode and already have 4 errors after 2 hours of testing. I really can't make head or tails of this, and the system event logs dont help.

UPDATE:

got an rma from the egg, and now have an asus a8n32-sli -- same freaking problem!! In fact, its worse on the asus board. I have tried 4 different sticks (2 kingston and corsair) of memory in single channel and dual channel, and in each slot. Memtest fails during the 6th test on each test round (about 5 minutes of memtest). At least on the Abit board, it ran correctly and never crashed in single channel mode. I am using default bios settings, this is fresh out the box. I have tried changing the memory voltage in each test as well (from 2.6 to 2.8), but I get hundreds of errors during test 6 in memtest regardless.

I would like update the bios on the asus board, but I am afraid of running into another problem. I can update it through the windows utility, but im screwed if the system freezes or gives me a bsod. I don't have the means to do it with a floppy. I saw another thread on AT, but the guy had a different asus board. He bumped the vdimm and that solved the problem, but not for me.


FINAL UPDATE:

BIOS in the Asus board was like 16 months old :\ Reinstalled Windows, updated bios, and now eveything seems to be working. Hopefully it stays that way.



System:
A64 4000+ (939)
Asus A8N32-sli
2 x 512MB Corsair XMS
EVGA 7800GT
Seagate 120GB SATA
Antec Neo550

 

[John]

[*]Since you have a total of 4 DIMM slots try the other two.
[*]Try cas 3-4-4-8
[*]Try 2.7-2.8v

 

[nervegrind3r]

I assume I can change the latency settings and voltage in the bios? Should I expect any problems since I am using Valueram? I have never oc'd and bumped voltage before.

I was thinking of trying the other two slots, but its another full week of testing since I run 24 hours straight for each test (6 tests). I may end up doing that.

 

[Insomnibyte]

The cas timings are mostly pre-set by alot of companies, and your motherboard uses preset settings unless specified otherwise. Valueram is just a universal part that runs in most systems. With dual channel you really want to make sure your parts are EXACTLY the same, this includes the build #. To find your build # its going to be directly on the moduel itself, and the sticker should have a 13 digit # that starts with 99. Should be something like 9905743-001.a00 or whatnot. If these #'s are different, than they will not work in dual channel mode together.
Btw, I work for Kingston Give me a call!

 

[John]

Originally posted by: nervegrind3r
I was thinking of trying the other two slots, but its another full week of testing since I run 24 hours straight for each test (6 tests). I may end up doing that.

That is excessive IMHO. If you can complete 2 passes of memtest86+, boot into Windows and run a 32M SuperPi and then 24 hrs of Stress Prime.

http://forums.anandtech.com/messageview...atid=28&threadid=1901991&enterthread=y

 

[nervegrind3r]

Well, the reason I test for 24 hours is because I get errors at random times. For example, when I started testing yesterday, I got 2 errors on the first pass, two errors on pass 6, and two errors on pass 8. When I checked my system this morning before I left, I was up to 22 errors. The 24 hours cycle will finish at 6pm today and ill probably have 30+ errors like I did the first time I ran a 24 hour test. Since the ram tests fine in single channel mode, there's something whacky happending in dual channel mode, even though the sticks are identical.

btw, I just updated the original post. Before I started running 24 hour memtests, I did run Prime95. It fails within an hour when using either option that tests memory (2nd and 3rd torture test), but runs fine when using the first test (that says Ram not tested much).

 

[Showtime]

RMA the ram. It will only cost you a little bit if anything.

Also, how stable are you voltages?
That ps should be plenty for your rig, but the if the power supply is not giving enough juice, there could be problems.

You can try manually raising the timings and see what happens, but it should all work at stocking settings. Is there another computer you can test the ram on?

 

[Swampster]

Check the manual on your motherboard. Many boards specify single-sided modules in (for example) in slot 1 and 3 and double-sided modules in 1 and 2. See what your manual advises and go from there. In the mean time . . . just turn off Dual Channel mode before something else gets corrupted.

 

[nervegrind3r]

thanks for the advice everyone.

I did verify that I can run dual channel in either slots 1 and 2, or 3 and 4 from the manual. The memory installation was correct for this mobo.

I just finished testing slots 3 and 4 in the same dual channel configuration for 24 hours, no errors at all. I was pretty sure the memory wasnt the problem, and this confirms it.

It seems there is some problem/corruption that occurs when running in dual channel in slots 1 and 2 only. Running in single channel, in slots 1 or 2, or dual chanel in 3 and 4 works perfectly. Im gonna RMA the motherboard, so I dont have to deal with this when I add another 2 x 512mb in the future.

this is the wackiest thing ive seen happen on any of my abit mobos. :

 

[Swampster]

Glad I could help you narrow it down.

 

[nervegrind3r]

OP updated, need some more advice.

 

[Swampster]

Check your motherboard manual VERY carefully. In many cases, you must run dual-channel sets in one pair of slots if they are single-sided (ie 1 & 3), and another pair (ie 1 & 2) if it is double-sided.

Also, check the recommended brands/models list to make sure your memory appears. Memory is NOT all the same . . . some will work great in most boards, but totally bomb in others . . . sometimes within the same manufacturer.

 

[nervegrind3r]

swampster,

thanks for the response. I verified from the manual that I am supposed to use banks 1 and 3 in dual channel, and bank 3 in single channel. I tried this, and got the same exact memory errors in either mode, and at any voltage between 2.6 and 2.8. When this failed, I tested each other bank and also got the same exact errors.

I checked the memory list in the manual. My kingston kvr memory is listed, but the corsair xms wasnt. I'm gonna check there update info on the website, im 99% sure ill find corsair xms listed there. In any case, all 4 sticks have the same memory errors during test 6 in all configurations I tried.

 

[Swampster]

OK, now I'm seeing something I didn't catch before . . . or maybe you didn't mention.

All memory must be of the same brand and model to work successfully in dual channel mode. Preferably, all modules should be purchased as ONE kit, as it doesn't take much of a variance to cause a timing problem when in dual-channel mode. That's why they call them "kits", because they are tested together to make sure they are truely identical.

There will always be variances in ANY manufacturing process. I remember years ago (I won't say how many <G>), when I worked around heavy machinery and many of them used multiple drive belts. One of the jobs of the maintenance department was to put belts under load on a special jig to find the ones that were not only the same length, but had the same deflection under load.

Not to be discouraging, but also keep in mind that you would have to have a really good stopwatch and a high-end game to find any performance difference between single and deal-channel mode with the amount of memory you are running.

Therefore, you might want to consider that, the thrill of solving a problem not withstanding, you have already spent more time and money than the benefits of your success would ever recover for you . . . just a thought.

 

[nervegrind3r]

the corsair is a kit, the kingston is not. I verified that the corsair is on the qualified list of memory as well.