Xtreme torture test of ram in Memtest86 (good for VX)

[starblazer]

I found this method of testing to be extremely useful in uncovering any potential ram problems that you might run into and want to bring this up to everyone to see if you guys think this is a good way to test.

Here is the situation:

I could run memtest #5 for 24 hours+ with zero errors, but with this localized section testing, I can bring up thousands of bugs in matter of minutes if the cooling solution in place is not optimal.

What I did was testing in small segments of 64 meg throughout the entire memory range. (I used version 1.40, don't remember if the older version have it, but I would assume it did) I would go in and set the range from 0 for lower and higher to 65536000 (64 * 1024 * 10) and loop test #5 for say 10-15 mins. The heat that this type of test generate on that section of ram is pretty extreme. You can run through thousands of test of this section of ram in matter of an hour. You can imagine the type of heat that this can generate over a regular #5 test where it move through the entire memory at a leisurely pace (in comparison) so the ram have time to cool off.

So, on my ram @260x9 2.4vdimm, I would be able to go through over a thousand loops of memtest #5 without error, but when I localize it to small sections like this, I would start getting errors as heat starts to build up to the point where it started generating hundreds of errors per loop after a few minutes.

You can then of course move the range throughout the whole 1gig of ram by overlapping say 4-8 meg. eg. 0-64 meg, 56-120 meg, 112-176 meg, etc.

Would this method be a valid method to test or is this too extreme for the ram that no normal computer usage will be able to generate this much heat for the ram?

[cantankerous]

It does sound pretty extreme but also very knowledgeable. Good job I think on this find.

[starblazer]

So anybody else tried this test to see if the extreme testing on local section of ram bring out errors previously undiscovered and thought to be stable?

[enzoR]

i thought test#7 or 8, i dont remember was an even more intense test than test 5 so maybe it produces simmilar results.

[cantankerous]

that would be test 8.

[STEvil]

Sounds like running test 8 on a localized amount would be interesting.

[craig588]

test 8 is just test 5 extended. It doesn't do anything more intense, it just repeats many times. Looping test 5 will give the same results.

[starblazer]

test 8 is just test 5 extended. It doesn't do anything more intense, it just repeats many times. Looping test 5 will give the same results.

Yes, that's correct. A normal run of any test in memtest is doing it throughout the whole range of memory, thereby allowing the earlier part of the memory to cool off before the next loop occurs. By, localizing it certain parts of the memory, the part that you are testing basically having very little time to cool off. This will allow the heat to build up over time that the usual loop test will not generate.

At the beginning of the localized test, the heat will ramp up, but it will not peak until at least 5-10 mins. But, once the heat reaches a certain temperature, it will not fall back down and thereby generate more and more errors. Even, with a fan blowing on it, it will not cool it enough. This is especially true if it's above 3.3v.

[viccyran]

I could only see this being useful for burning in your ram more quickly, I don't think there is much of a stability proof for this, as no program that you'd actually use requires the use of one specific region multiple times quickly

But a nice burn-in I must admit, running it to fully burn in my friend's BH-5 that he just discovered he had...

:toast:

[Rabbi_NZ]

really good find dude... there is always a certain IC on a memory stick that maxes out first... this way we can burn-in the poorer chip only, hopefully, strengthening the module's weakest point.

[starblazer]

I could only see this being useful for burning in your ram more quickly, I don't think there is much of a stability proof for this, as no program that you'd actually use requires the use of one specific region multiple times quickly

But a nice burn-in I must admit, running it to fully burn in my friend's BH-5 that he just discovered he had...

:toast:


That is definitely the question I have in mind. Will any program actually use ram in this way? The one thing I do know is that with my VX ram at 250x10 @ 3.3v, I can run this localized test for hours without any errors, but at 3.4v, the heat overpowers any fan I put blowing on it and you can feel the heat from that section of ram.

[starblazer]

really good find dude... there is always a certain IC on a memory stick that maxes out first... this way we can burn-in the poorer chip only, hopefully, strengthening the module's weakest point.


Thanks. In fact, that's why I started doing localized burn in because I was always getting errors in a certain memory range. (eg. 610-640 meg). So, I just totally stress test that section of ram. The errors did in fact go down after some extreme burn in.

[cantankerous]

oh sweet. I too have been having errors in the same range. Usually 400-500 range. Perhaps I can burn them out with your method. I'm going to give it a try. Thanks.

[Tony]

guys

We really do not think this is wise.

Reason....to much heat is generated in a localized test, this could damage the IC's.

Better to be gentle than brutal ;)

[Rabbi_NZ]

Thanks bigtoe :toast:

Does OCZ have a theory on what level of voltage is best for burnin?
Reason I ask is the best BH5 module I ever owned was one that had been run at 2.8v for over a year at mild overclock....

[starblazer]

guys

We really do not think this is wise.

Reason....to much heat is generated in a localized test, this could damage the IC's.

Better to be gentle than brutal ;)


Thanks for the response. On that note, wouldn't it be dangerous if a localized memtest like this can be dangerous? Wouldn't it be possible that a virus can be written that it's specifically target this weekness? Even if you run way below the warranted voltage, can it still be dangeorus?

[Tony]

I just use test 5 and work the fsb up a notch at a time and slowly apply voltage. I posted my method a while back and while being slow has helped me get a lot of BH5 to ddr500 2-2-2 before...thats why i used it on VX.

Always best to burn in slow...my theory anyway.

[Tony]

Thanks for the response. On that note, wouldn't it be dangerous if a localized memtest like this can be dangerous? Wouldn't it be possible that a virus can be written that it's specifically target this weekness? Even if you run way below the warranted voltage, can it still be dangeorus?
The issue is your running high voltage and a test that generates a huge amount of heat with no let off...the ram just gets hotter and hotter.

Lets keep virus talk to PM's and give no one idea's

[[XC] moddolicous]

I tried to figure out how to do this with Memtest 86 version 3.2, but I cant. Can someone tell me how to do this?

[[XC] moddolicous]

can someone run me through the steps?

[craig588]

get memtest+

The best section to loop would probably be the area between your highest banks, like 500-520 for 3x256M single sided sticks of ram, or 760-780 for 2x512M double sided sticks of ram and so on. Bank switching is apparantly really hard on the memory controller. (Or at least on mine, it get errors right around those areas if I take the speed too high, it doesn't matter what order the sticks are in)

[[XC] moddolicous]

Do you mean memtest 3.2. I cant find a way just to run one single test repeatedly. Can you tell me how?

[Rabbi_NZ]

Do you mean memtest 3.2. I cant find a way just to run one single test repeatedly. Can you tell me how?
C - 2 - 5 - "Test Number" - Enter - 0

[[XC] moddolicous]

thank you, I'll report back with results.

[ryanpgroovy]

that quite scary as you could potentialy kill the chips by localizing the heat build up .


I really think its wise to keep the chips cool during any " burn in"

[conrad.maranan]

I agree with Ryan. Here's a safer way to do it:

- Find the maximum frequency you can loop Memtest Test #5 without errors
- Tweak the voltage for the above frequency until you start noticing errors in Memtest
- Loop Memtest Test #5 at the above frequency and voltage overnight
- Loop Sandra Memory Burn-In the next night
- Loop 3DMark2001 Lobby the third night
- Loop Prime95 Torture Test Blend the fourth night
- On day 5, see if you're stable and error-free at a higher frequency.

I've got it in my congested head that the above will work well because you're making good use of all memory addresses.

Believe it or not, I've had my VX for almost a month and I'm still burning it in in hopes of accomplishing better when I step up to NF4/FX-55.

Good luck! And keep those fans blowing.
:toast:

[Rabbi_NZ]

One more thing... once you find a Memtest #5 stable point I advice running MemTest on ALL tests for a night or so. You may find other tests are failing... for me, test #7 seems to indicate the memory controller needs more vCore.

[andyOCZ]

I have to weigh in also. Any testing that is cuasing exterme heat can be bad. Real world testing in much better.