A7N8X-E Deluxe Dual Channel = Bad RAM? (or two modules = bad RAM?)

Testing memory exhaustively is an incredibly slow process, so I haven't had a
chance to troubleshoot this _fully_ yet.

I'm running an A7N8X-E Deluxe with a "TwinX" pair of Corsair 512MB XMS PC4000
modules (250MHz/500DDR RAM running at 200MHz/400DDR, and with default
timings; nothing aggressive).

Has anyone ever heard of nForce2 motherboards having trouble running
error-free in dual channel mode? Or has anyone ever heard of two modules of
RAM being more error prone than one?

I just downloaded Half Life 2 from Steam, and it's crashing like crazy. I
can't play for more than a minute or two, it seems. I've had periodic
crashes before (two per month, maybe), with IRQL_NOT_LESS_OR_EQUAL errors,
and one with PFN_LIST_CORRUPT (WinXP BSOD errors). And the World of Warcraft
open beta was crashing on me regularly, too. Most other things work fine,
and I can usually run for quite some time with no errors. I've also found
that if I run various memory testing programs for long enough, I invariably
find an error. It's not repeatable in a given location, but I am getting
intermittent memory errors. I suspect that's the source of a lot of my
crashing problems.

So last night, I took a memory module out and ran MemTest86+. I told it to
run all tests, but unfortunately then told it to restart the test (thinking
that might be necessary to have all tests kick in), not realizing that doing
so resets the settings, and it goes back to its "quick test" (which still
takes quite awhile). This morning, I got up and checked my computer - It had
gone through the tests 14 times and found no errors (previously, I had found
errors within the first 2-3 passes using the same quick test and both RAM
modules). So I swapped out the module - putting the other module back into
its original slot, and taking the first chip out. This way, I figure I could
narrow my problem down to a bad module OR a bad slot, and then narrow it
further from there.

This time, I ran ALL tests - after 2-3 hours, it was still finishing up the
first pass (92% done or so). But so far, no errors. I'm going to keep
testing (I needed to reboot into Windows to get some information, and figured
I'd write this email while I did), but right now, it looks like NEITHER chip
is bad, and NEITHER slot is bad, but the combination is bad. Is that really
possible (or likely, for that matter)? If so, is the problem likely to be
solved by moving the modules around such that they run in single-channel
mode? Either way, is the fault with my Mobo or my RAM?

 

<<snip>>

The Nforce2 is picky about RAM, and I consider the RAM thing to be
the biggest exposure now with using Nforce2. I had some RAM that I
thought was fine, but in a new A7N8X-E, it would work fine in dual
channel, only up to 180MHz. In single channel mode (two sticks in
slot 1 and 2), it would work up to 200MHz. My theory is, there are
too many I/Os for the number of power/ground pins on the Nvidia
Northbridge package, and the I/O timing slows down when both channels
are engaged.

Memtest86 would pass the memory, so was useless for debugging the
problem. I found Prime95 from mersenne.org was much better, and
it would detect an error in my case, within 30 minutes of testing.
(Use the "torture test".)

My solution was to buy some Ballistix PC3200, but I really don't
know what aspect of RAM (clock speed or low latency) the Nforce2
is really looking for. It is obvious in your case (and another case
I read about where someone was using PC4000 memory), that clock
speed alone seems not to be enough to fix it. Maybe the board is
looking for low latency memory ?

I'm having trouble correlating the symptoms, with
how the memory works. SDRAM and DDR are synchronous technologies.
They should only be sensitive to meeting Tsu and Th (setup and hold
time), and not be sensitive to CAS latency. In other words, if
normally you had a memory which was acting up, increasing the
CAS setting by a notch (CAS2.5 to CAS3, say), would be enough to
make a memory stable. That is a case where the internal timing is
failing, and the CAS change solves that internal problem. But in
the case of the Nforce2, that doesn't seem to help. In your case,
you are using a high clock memory, and such a memory should have a
fast clock to Q timing. But that isn't helping you. So, I'm puzzled
as to exactly what is failing here. If the failure was external
timing, then a memory capable of being clocked faster, should
have helped, or at least exhibited different symptoms.

I hate recommending a purely random purchase of RAM - you can
visit Nforcershq and see what some of the users over there
recommend, but I'm not sure that memory DIMM construction stays
constant long enough, to use such a buying recommendation
successfully. A DIMM maker can change chips at any time, which
nullifies the benefit of shopping by brand.

You've tried a high clock memory, so the only other type to
try, is a low latency memory.

HTH,
Paul

 

I'm actually finding that this isn't a dual-channel problem, but a
dual-module problem. I put the modules in the first two slots to force it
into single-channel mode. While both chips individually check out fine, I
get errors with both of them in single-channel mode. I'm not sure that the
errors were as frequent as they were in dual-channel mode, so it remains to
be seen whether or not they're enough to cause problems with any frequency.
But if it does cause me real-world usage problems (I suspect it will), it
looks like I may be looking into new memory. I bought Corsair XMS because it
was supposed to be among the best memory out there, and figured that as
finicky as the nForce2 boards are known to be with memory, that something as
reputable as Corsair should work fine. Guess not.

I haven't run Prime95 in awhile, but when I first got this system set up and
started seeing very occasional crashes, I did run Prime95. It had no
problems.

I was hoping to find a workaround that didn't involve buying another gig of
RAM. But if that's all I can do, I guess I'll do that, and put these
modules up on eBay, since they're perfectly good memory modules.

 

I found my solution. After looking around, I found a Corsair help forum -
apparently these memory modules need more power than the motherboard
allocates to them by default. The Mobo defaults to a 2.5V DDR reference
voltage, and the Corsair XMS chips need 2.8! I upped the voltage, and that
seems to have done the trick. Now I just need to put them back into their
original slots to get them back in a dual channel configuration and see if
that works.

 

I wish I could recommend something, but some of the stuff that works
(like Winbond BH5 chips) aren't made any more, and what is really
required, is a memory poll based on sticks that are currently
available.

http://nforcershq.com/forum/viewtopic.php?t=2482

(Memory poll)
http://nforcershq.com/forum/viewtopic.php?t=7824

Perhaps you could contact Corsair on their help forum:
http://www.houseofhelp.com/v2/forumdisplay.php?forumid=128

Maybe they can be convinced to swap the PC4000 for something
known to work well on Nforce2. (3200XL?)

Have you tried bumping up the memory voltage ?
Just enough to see if the error rate responds to a voltage
change or not.
http://www.houseofhelp.com/v2/showthread.php?t=32391

http://compatible.corsairmemory.com/Default.aspx
I guess they think Nforce2 works with just about anything...
http://compatible.corsairmemory.com/memorysearch.aspx?modelid=461

They actually claim to test with A7N8X here:
http://www.houseofhelp.com/forums/showthread.php?s=&threadid=8048

HTH,
Paul