underclocking cures memory faults? help diagnose

[Guest]

MACHINE:
Piii/866, mb CUV-4XE, 128+256MB PC133, no ECC, multiplier locked at 6.5.


SYMPTOMS:

IF FSB=memory= @ 133 or 125 MHz (so CPU @ 866 or 812 MHz)
THEN
* 100% reproducible crashes at boot (WinME, and 2 different Linux distros),
* memtest86 reports 10's of 1000s of errors already in Test 1.

However, IF I underclock further by setting
- EITHER FSB @ 120MHz (CPU @ 780) + memory @ 100% of FSB (i.e. @ 120)
- OR FSB @ 133MHz (CPU @ 866) BUT memory @ 3/4 of FSB (i.e. @ 100)
THEN
* system stable
* memtest86 runs Test 1 fine
* and I can write this to you on the affected machine...


QUESTION
Is this likely to be a memory problem,
or do I have to suspect the northbridge may be roast?


Will appreciate drops of wisdom!

Spammy

 

[Will Dormann]

MACHINE:
Piii/866, mb CUV-4XE, 128+256MB PC133, no ECC, multiplier locked at 6.5.

QUESTION
Is this likely to be a memory problem,
or do I have to suspect the northbridge may be roast?

Likely a memory problem. What if you use only one module to run
memtest? (try testing each, individually)



-WD

 

[Guest]

I will! Thank you!

Spammy

 

[larrymoencurly]

IF FSB=memory= @ 133 or 125 MHz (so CPU @ 866 or 812 MHz) THEN
* 100% reproducible crashes at boot
* memtest86 reports 10's of 1000s of errors already in Test 1.

However, IF I underclock further by setting
- EITHER FSB @ 120MHz (CPU @ 780) + memory @ 100% of FSB (i.e. @ 120)
- OR FSB @ 133MHz (CPU @ 866) BUT memory @ 3/4 of FSB (i.e. @ 100) THEN
* system stable
* memtest86 runs Test 1 fine
* and I can write this to you on the affected machine...

Is this likely to be a memory problem,
or do I have to suspect the northbridge may be roast?

I had the same problem with some Kingston and K-byte PC2100 modules
and two different types of mobos (P4 and Athlon), and slowing the
memory bus from 266 MHz to 200 MHz was the only thing that eliminated
the errors. Even changng every other memory parameter to its slowest
setting didn't help. All the bad K-byte modules had Spectec chips
(recycled Micron), while the good ones had Elixir chips (recycled
Nanya), which worked fine even with all but the most aggressive timing
settings. All of my Kingston modules had no-name chips, and while the
old US-assembled ones were fine (but not great for overclocking), the
China-assembled and new US-assembled modules consistently failed
except at 200 MHz. Kingston tech support was useless and couldn't
explain the differences among their modules -- they said that the old
ones were exactly the same as the others/very different from the
others, depending on who answered.

 

[Guest]

(e-mail address removed) (larrymoencurly) wrote

I had the same problem with some Kingston and K-byte PC2100 modules
and two different types of mobos (P4 and Athlon), and slowing the
memory bus from 266 MHz to 200 MHz was the only thing that eliminated
the errors.

There is a silver lining, especially if the memory modules did work as
advertised for a while: you can hope that a failing module may work
well for another while, once "derated".

I know this is exactly what CPU-makers do: they use a certain design
and process, but sell the "same" chip at different reliable,
marketable specs after testing, and apply continuous process
improvement to raise the spec mix of the product population.

{ Does anyone know if this is the case for RAM as well? }

This is my first memory "repair", but I think I was methodical enough.

I found that the bad mem was the older 128MB stick. This is what I
did:
- got a memtest86 v. 3 boot disk
- tried one stick at a time
- kept alternating between memtest86 and BIOS settings page
- started from deep underclocking and worked my way up
- once one stick started failing, I kept going higher using the other
one alone
- I eventually stepped over spec (133MHz in this case) and did a whole
11-test cycle at 140MHz.

Once satisfied I had a culprit, I
- went out and bought a 2nd hand replacement stick, cheap but branded,
from the PC shop between the local delicatessen and cafe'
- stuck it in and briefly tried the machine at 140MHz, successfully
- then reset the memory rate to 133
- bunged the "bad" 128MB stick in an older comp...
-...which just completed its 4th full 11-test memtest86 suite at
100MHz, with no errors.

After several days at 100MHz on my old main computer + 3 hours
churning tests on the even older one, I think this poor bad stick
earned itself a career change!