weird stability problem/memory errors, experts needed

Hello,

I've been fighting with a difficult to pinpoint stability problem for
several months, and it's gotten worse recently. Here's the story:

A few months ago, I bought the following config:
- AMD XP 2000+ , Thermaltake fan
- A7N8X Deluxe, 1002 bios or so
- cheap PC2700 RAM

I immediatly ran into stability issues under Linux (no Windows on this
machine) : unusual and random program and kernel crashes, which made me
think about memory corruption.
This was confirmed in memtest86, which gave dozens of memory errors,
with *always the same bit mask* that got copied/read wrong. So it looked
like some pin on some bus was fried somewhere

Of course, I removed all unneeded cards and peripherals, updated the
BIOS, tried all kind of BIOS settings, DDR voltage, mem frequency, etc,
to no avail.
The CPU heat was something like 50°C, pretty much ok.

So I swapped the RAM chip for a Samsung PC2700 one.
No better luck. So it wasn't the RAM.

I then tried putting an XP1700+ in, and if I remember correctly the
memory errors disappeared. I thought that it was the motherboard somehow
failing at the bus frequency required for XP2000+ operation, but never
believed it could be the CPU that failed.

I brought the whole config to the shop, where they separately tested
each component and diagnosed the CPU to be the culprit, causing random
3D Mark crashes.

So far so good, the CPU was replaced for another 2000+, and off I went
merrily for a few months, until...

2 weeks ago, I noticed my HD's were running too hot (50°C) due to poor
air flow in my old case, so I decided to get a new shiny case with
proper cooling (3 case fans + power fan).

The HDD's temperature got much lower (like 40°C), and so was the overall
case temperature, so I decided to slow down the CPU fan a bit. The max
CPU temperature was still around 50°C


But amazingly, a week or so later, I started getting those stability
problems, and they got worse every day.

Yesterday, I launched memtest86 again, and *gasp*, I get THOUSANDS, not
dozens of memory errors, with the same error bitmask as before.
"****, did I really fry my CPU again with so little fan tweaking ?"

I updated to the 1004 BIOS, just in case, to no avail.

So today, I went and bought a new FSB333 XP2600+ + low noise Spire
Coolers fan, and crossed fingers that it'd solve the problem.

Well, no. First, the A7N8X kept the 133Mhz bus setting from the old
2000+, but I didn't notice it, and just launched memtest86 which gave
the thousands of errors again. This was seconds after boot-up, with a
properly mounted CPU fan, and a 45-50°C CPU temperature.

I then set the bus setting to 166Mhz as it's supposed to be, and tried
again "optimal" and "aggressive" settings, but nothing could make the
errors disappear.

So now I've run out of ideas, I'm rather depressed, and wondering if I
should just go out and buy another A7N8X + some PC3200 chip tomorrow.
But of course, that ain't the cheapest solution, and I'd like really
like to find the real culprit now...

So, any expert ideas out there ?

Thanks A LOT!

--
Renaud Guérin
http://renaudguerin.net