I've got a p4 2.8ghz HT processor running on an Asus p4p800 Deluxe mobo
with a pair of pc3200 256mb Corsair Twin-x dimms.

The system has been unstable of late, so I hauled out memtest86 and
gave it a run - sure enough, a few errors showed up, each of them in
one of two locations, both around the 480 mb address, as memtest puts
it.

So I took out the second dimm, ran memtest again, and got no errors.
Then I tried swapping the positions of the two dimms from their
original slots and ran it again - errors again, in the same addresses
as before.

Finally, I tried running the test with just one dimm in the second slot
- no problems.

What do you suppose this means? I've been trying to sort it out - it
seems like there's some problem handling more than 256 memory, but it
doesn't appear that the slots have anything to do with it.

Do you suppose this means my motherboard is (slightly) cooked, or is it
the ram? I don't see how this could be processor-related, but it's not
outside the realm of possibility.

Anyway, if anyone more hardware-savvy than I has any thoughts on this
subject, do tell - I'd be much obliged. Perhaps someone will be able to
tell me exactly what the problem is from my meagre description, but I'd
settle for some advice on how to further isolate where the problem is.

Thanks,
Jason

First I would check the BIOS memory settings
and make sure that they are at the defaults.
Many, if not most, mobos specify slower memory timings
for 2 DIMMs than for only one. If no joy there when
testing the 2 DIMMs, I would the run
memtest on each DIMM separately in slot A and then
again on each one in slot B. Let it run for a long while.
If no errors show up, then I would conclude the problem
lies on the motherboard.


"Jason" <jasonsewall (AT) gmail (DOT) com> wrote

Previously Jason <jasonsewall (AT) gmail (DOT) com> wrote:
Likely one of the modules is operated at its limit in single-module
configuration. (That the address does not change when you swap modules
does not necessarily mean anything. The BIOS could "sort" the modules.)

With two modules the bad one is over its limit. Some signal-lines
are shared and the additional module puts capacitive load on it,
so the timing gets a bit tighter. In addition memtest86 cannot
reliably identify problems, since these ofteh only show up with some
probability. I recently had a bad module (Infineon) that only
caused one error in about 20-30 test-cycles of memtest86. It did
cause frequent problems in some scientific data analyses.

I see two choices for you:

1. Try to make the RAM timing slower and see whether the problem goes
away.

2. Try to get replacement modules from Corsair. Tell them what you
observed and ask whether that is enough for them to replace the
modules.

In any case, also when you try to identify the defect module, run
memtest86 at least for a day.

Arno