Passes memtest86, fails prime95

I have a question for all you memory vetrans.

I have 2x512 sticks of Mushkin 222 special, which I have to run at 2.8v to get stable. Thats fine. Mushkin said they will still warranty them at 2.8v. Here is whats wierd. My system appears to be rock solid stable, but the In Place Large FFTs test in prime95 will eventually error out after a 1-2 hours. I have been tearing my hair out to figure out why. I ran memtest86 just fine so I had initially ruled out my ram.

Then on a hunch yesterday, I pulled one stick out and ran the Prime95 torture test it ran for 8 hours or so, no problem. Then I swapped the sticks and it failed again after two hours. Now I am back to the "good" stick, testing to see if prime95 will run again for 8+ hours.

My question is, how often does memtest86 not detect bad ram? I am going to have to warranty the one piece I think, but I was suprised that it seemed fine in all but one circumstance/test.

I had some KHX3000 that did the same thing. I could Memtest all the way up to 215 MHz or so. Prime would miserably fail all the way down to 190MHz no matter what I did with timings or voltage. Windows, games, and 3dmark all had probs with anything above 190 as well. Pulled the spreaders and these were the non BH-5 variety. Memtst is obviously not an ends all when it comes to finding bad/questionable ram.

The 222 special that I have now will pass memtest up to the same speeds as well. At 200 MHz at 2.8v they are fine in windows, games, and 3dmark. No instability there. Prime is the same as your situation. I will eventually get warnings after a while, but no errors.

Hmm, that is odd. I suppose that memtest86 just cannot test every possible combination of data patterns, but it does test a good representative sample of them.

One question: when you ran with just the "bad" stick, did prime95 run into trouble right away, or did it take longer? Did you use the same RAM slot as that used for the "good" stick?

How are your memory voltages holding up under load? Is your memory or memory controller heating up too hot after extended use?

Does the 2-stick combo pass multiple runs of memtest86, or did you just do one iteration?

I've seen similar issues with mine, although I haven't had a chance to test extensively, where my 3 sticks work well for 1-2 iterations of memtest86, but start giving errors after a longer running time. Not really sure what that's all about.

Just thought I'd toss some ideas out to look into. -- Paul

It might be that I didn't run memtest86 long enough to really get the errors out. I ran both sticks in dual channel through two iterations w/ no errors. I don't really like using memtest86 becaues I can't do anything else on my machine while running it.

The two stick combo will error in 1-2 hours in prime95.

The single "bad" stick will error in 1-2 hours in prime95.

The single "good" stick, in the same slot the "bad" stick was in, seems to run indefinitely in prime95.

The memory does get pretty toasty, hot, but not untouchable. Haven't checked the memory controller.

I am becoming thoroughly convinced now that the problem is just the one stick. I will be calling Mushkin to see if I can warranty it.

im not as obsessive as some are. u might try running ur apps gamimg etc and see howthey do.. if u dont get errors or defaults to windows while gaming... i wouldnt worry.

@zbose:
Hmmm, it's defintely sounding like it's the one stick. Remind me: did you say the errors occur at stock speeds as well? Did you say it occurs in the same slot?

@flapperhead:
On the other hand, if you do work with your computer (such as large computations or simulations), you definitely don't want memory errors. An error in a game is no big deal (a misplaced polygon or AI gone nuts), but a memory error in a CAD application or during a structural engineering stress simulation ... -- Paul

This is at stock speeds, which is why it worries me more... it doesn't bode well for overclocking. Before I start increasing anything, I would like to know that I had a rock solid stable system at stock speeds. Besides, as the stress.txt file from prime95 says: why test for it if your just going to ignore the results? That and it will drive me nuts to know that my system has a hidden flaw like that! Maybe I am a wee bit obssesive.

macklin, in the same slot, one stick runs prime stable, the other does not. And this is all at stock speed, 200mhz.

Well, if it errors at stock speeds in the same slot that works for another stick, then I think we can rule out the memory controller, PSU, and any other issues. I think it's time to look at getting that stick replaced.

It takes a lot of discipline to track down these errors at stock speed before proceeding. You're doing a great job. Thanks for answering my questions. -- Paul

Just wanted to post an update to everyone. Got the second stick back from mushkin (a new one, different color heat spreader! now they dont match!). So far, running 24 hours + on prime95. Looks like im stable!

Now to reduce stability by overclocking.... hehe.

Also, question for anyone who may read this, if I can run at 2-2-2-5 should I? I know, I know, benchmark it and see... well I don't ever seems to be able distinguish much of a difference! Even between dual-channel and single channel...

hrm.

I've had settings where it passed about 10 hours of memtest then failed about 13 hours of prime95.