Technical > Overclocking / Modding > memtest86 conditioning

reply to Exit

Re: memtest86 conditioning

said by Exit:

---

I found that if prime 95 is stable then so is your ram. Since it does math calculations it is a really good way to check for errors

Steve

---

reply to paul
Thanks Paul!

Just in time for me to put a new cruncher together this weekend and put her through the paces.

Did the .iso to CD version, works great!
--
"I have arrived at a positive, definite, unchangeable, inflexible, conclusion. That is, if I do not change my mind." Groo. :Þ

reply to paul
Hmmm, back at it at again, trying to get the Corsair to pass Memtest86 at 2400+ mhz, Barton 2500. In test 5 looped, I'm getting 5k errors at 98% done. Clean all the way through to 88% done, then it jumps to 98% done, spewing errors. The same stick does 216-17 fsb error free when the multi is dropped, running inthe 1800mhz range. I'm down to 206 fsb, still looking for that error free zone. I know this was explained once before, but don't recall where.

Paul
--
looking for little green men.... Team Starfire Queue VI

memtest #5 is really two tests. That last 10% involves relocation of the program in memory so that *all* the memory can be tested. So, not just the memory is tested, but the ability of the system to relocate a running program is being tested. The range your test is failing lies in that relocation range.

Try turning off the cache for test 5. If test 5 then completes that last ~10% without errors it points to the CPU being the culprit. If test 5 then completes that last ~10% WITH errors it points to an overall system instability (the memory/memory controller/cpu subsystem).
--
Don't confuse me with the facts...

reply to paul
Ok, I ran memtest86 looped test 5, all mem range, and turned off cache (in memtest options), still getting huge amounts of errors in the last 10%. Backed off to 11X200, still getting loads...

What should I look for? Heat? Too much vcore or vdimm?

Just backed off vdimm to 2.77, errors dropped dramatically @ 11X200, cache disabled. Not much of a difference when enabling cache, too.

Trying vdimm @2.63, still 11X200, down to 264 errors in the last 10%, cache enabled. Cache disabled, roughly the same.

edit- while on the temp theme, realized I was running vcore still pretty high. dropped it to 1.8V, no errors. Now at 2200mhz, 11X200, cool and error free, cache mode on or off!

Paul
--
looking for little green men....
Team Starfire Queue VI
[text was edited by author 2003-05-25 19:02:43]

I had similar problems conditioning Samsung PC2700 DTL.

Keeping in mind that memory transfers (test #5) involve the CPU, the controller and the memory I decided to lower all timings to where it went through test 5 without errors. I looped test 5 for 12-24 hours then tested higher settings.

Eventually, I conditioned the memory sub-system to the point that, from a starting point of 180 FSB without errors I was able to run test 5 at 200 FSB, no errors.

Tuning and conditioning. I know you don't have as much time to do it as did I, but that's the only advice I have to give you...
--
Don't confuse me with the facts...

reply to paul
Almost to the point of having a rig just for conditioning.

Really sunk home with this exercise how much heat affects overclocks.... I've got all in ones running close to stock running pretty hot in a storeroom, they'll run forever like that. But ramp one up, you need the temps to be in the low 40's board sensor or mid 50's diode temp.

Amazing that the boxes will run fairly stable even with memory errors like I had above.

edit- just happened across some other people at AMDMB with similar problems with the Barton. Wonder if the big L2 cache is prone to heat problems?

Paul
--
looking for little green men....
Team Starfire Queue VI
[text was edited by author 2003-05-25 19:54:05]

After thinking perhaps the heat in the L2 cache with a decent overclock at high fsb was causing instability, I swapped the Barton out with a 2400+

The Barton could pass memtest at 2150-2200mhz, with a vcore of 1.8. Anything higher would cause errors in memtest.

The 2400+ is running 11X210, 2320mhz, 1.85 vcore currently with no errors. Will keep trying combos to see what it will do.

Paul
--
looking for little green men.... Team Starfire Queue VI

Actually thought it might be the Barton cache, even started a post to say that - but - not having a Barton I didn't want to guess, even if it were an "educated" guess lol...
--
Don't confuse me with the facts...

reply to paul
Certainly seems so. Now up to 2413mhz error free 11.5 X 209.

Need to drop the multi and see how high I can get the fsb stable. Tired now.... Thanks for the insights!

Paul
[text was edited by author 2003-05-25 23:13:33]