So I've been winding my E6750 up. Got it seemingly nice and stable at 3.2. (FSB 400, DDR2-800).

However, as I'm benching today, SuperPi keeps hanging. Finally I get it through and make a note to run Memtest86.

So I get the latest Memtest86 (3.something), and it keeps bombing about 13% of the way through test 3.

The error is so consistent, and the message so bizarre (basically suggesting an error somewhere in the lower 2Gb of address space, real useful), that I figure "maybe the disc was bad".

I try Memtest86+ 1.70.

I start getting a lot of errors, mostly tests 5, 6, and 7.

I'm testing now at 2720 (which is where my mainboard has decreed "stock" speed is (FSB 340, memory DDR2-816).

I run the sticks singly, and one stick clears test 5 fine, the other tends to produce about 60 errors in test 5, then a few more in test six. This at voltage of 1.85.

Sounds like that stick is bad.

Here's where it gets weirder:

You might figure "maybe it's not getting enough voltage". After all, the system won't even boot reliably if you don't set the RAM voltage to 1.85. I discovered this on day one. So I try juicing it with that one stick (setting it to 2.05, the stick says "1.8-2.2v" right on the label!) I actually get more errors than before-- errors in test 1 onward.

So I've pulled the stick out, and now have 1Gb of single-channel DDR2-816 and a stock-speed setup.

Discussion questions:

1. Could overclocking the CPU have somehow damaged the RAM? Considering the timings were never above DDR2-816, I'd expect not. The fact it passes test 5 with the "good" stick alone makes me assume that it's not a CPU or mobo issue.

2. Is this a RMA situation? My gut is saying yes.

3. If it goes to RMA, how bad is this going to be? If I get on my bike now, I might be able to clear the Newegg return window, but even still, I feel like there's going to be a problem because one stick seems to work fine.

4. What about mismatching pairs? If I'm going to have to spend $25 for insured postage, or send back both sticks and be out of commission for weeks, I may as well just eat this stick, make a funky keychain, and buy another 1Gb stick locally. Or is that a recipe for disaster with dual channel? The only time I ever tried it was with two 256M DDR266 sticks (one Crucial-with-Micron-chips, one Crucial-with-Samsung-chips) and it was part of a "it acts like the memory, but the PSU is really failing" situation.

The stick in question is labelled:

1GB PC2-6400
DDR2 800 CL=5-5-5-15
VOLTAGE= 1.8-2.2V
GX22GB6400DCK=
PN: 105ST6DS2566B
ASSEMBLED IN USA

(evidently there's the problem right there )

5. Is there any other knob that might just be set funny by my board that could have triggered this problem? I have the memory timings at "By SPD", and CPU-Z reports (on the presumably equivalent working stick) timings of 5-5-5-15. Of course it also reports a voltage of 1.8v is intended by the SPD, and we all know how well that worked.

If it were me I would try and RMA both sticks to stay with a matched pair. In the meantime buy you some cheapie stuff for backup ram.

I actually have some cheapie ddr2 (2x512) I would trade for or let go cheap. PM me if you like.

Incidentially, does anyone know if memtest86 3.4 is funky in some way?

That was my first stop on the diagnostic journey. It kept bombing at test 2, 13% or so.

It also does so with the single stick that passes Memtest86+ 1.70

and more interestingly, it bombs at test 2, 20% or so, on a second system believed good (X2 4600+, four 512M sticks of Corsair Value-Select DDR400)

I find it highly unusual that two very different systems would fail the same diagnostic in very similar ways. I'm going to grab another copy and see if I had a bad ISO.

However, this would not change the fact that trusted reliable Memtest86+ doesn't like one stick of this memory.

If higher voltage leads to more errors I would assume there is a physical defect in the ram. Higher voltage would exacerbate an electron leak, or bad solder joint.

Well, this story got a lot weirder.

I bought another 2Gb set, and the stick in slot 3 was again bad (when tested alone in slot 1, it failed).

At this point, I'm suspecting that either I have monster bad luck (like 10,000:1 against odds) or the board is eating RAM, so the board's gone. I got a Gigabyte DS3L and things seem happy now.