It's not quite OC, but voltage fiddling is in the same general territory.

Here's the memory:
New: Two sticks, Patriot DDR2-800, 2Gb each, 5-5-5-12 timings, and 2.0v rated
Old: Two sticks, Corsair DDR2-800, 1Gb each, 5-5-5-18 timings, and 1.9v rated.

Here's the platform:
Gigabyte P35-DS3L PCB version 2.0, BIOS version F7. The memory voltage settings are (per the BIOS hardware monitor):

"Normal" = 1.90
"+0.10v" = 1.90
"+0.20v" = 2.02

On top is a C2D E6750 at stock, and an Antec SP-500 PSU.

Here's the stability rundown:

-At standard voltage, Memtest86+ runs successfully (one test at a time).

-So I put it down for the night, and fire it up first thing in the morning to track a package. I get a hard freeze (mouse just stopped, everything just stopped, hard reset required). Looking at the logs, all I get is "The previous system shutdown at 8:09:17 AM on 2/19/2008 was unexpected.". Forty seconds earlier is the last log from the prior boot-- a mention of updating the Windows Update components. Maybe I panicked and it was doing some bad CPU thrashing that froze up the mouse, a PSU voltage sag, or just bad luck?

Recalling that SuperPiMod is an excellent stability test, bombing with Memtest86+ won't, I run it, and notice it bombs. A lot. All the time. What I didn't realize at the time: Since I switched to Vista-64 to accomodate all 6Gb, I was adding new complexity. SuperPiMod HATES Vista-64 apparently. (Super PI with msec/anticheat mod - Page 9 - XtremeSystems Forums )

I begin tweaking the voltages (+0.10, then +0.20), and continue to get SuperPiMod failures, just getting more panicked until I discover that it fails for every Vista-64 system.

I'm hoping 15 minutes of the higher voltage won't ruin 1.9v memory.

Learning of the SuperPiMod problems, I reset to stock voltages, and it hasn't frozen again yet (crosses random appendages, uptime is 30 minutes), but I'm not sure what to do.

If I go 2.0v permanently, the Patriot memory will surely be happy, but the Corsair might barbecue. I don't want to just waste 2Gb of memory for no good reason. There's no "split the difference and run at 1.95v" setting. Any thoughts? Roll the speed down to DDR2-750 or something and hope it can reach that at 1.90v with 1000% stability?

One other thing-- I noticed this memory is very strong with the "New electronics smell". When I first ran the whole 6Gb in Memtest in an open case, you could definitely smell it. When I jumped to 2.0v, I got another whiff (less strong, but the case was closed). Oddity?

Your memory should run fine at stock volts unless your clocking them higher. Odds are you can run at faster speeds if you go to 2.2 volt's. That what the recommended volts are for the faster memory.
What volts are you running on your cpu? I would suspect that you need to up that voltage.

I seriously doubt your ram will cook at those volts. As long as you have decent case airflow you should be fine with 2.0

It sounds like I have the same Corsair XMS2 DDR2-800 ram. I've had mine at 2.0v and it's doing just fine.

I suppose if I get any more irregularities, I'll bump to 2.0; at 1.9 everything *seems* to work surprisingly well, and it probably runs cooler. I found it ran damned hot when I tried four sticks installed before and ran it heavily (Memtest86+)

[edit] And lo, the irregularity occured. Another hard freeze, strangely also in close proximity to Windows Update (I was refreshing it to see if SP1 was being pushed to my box yet), so I'm at 2.02. I hesitate to fire up Memtest86+ and see what happens [/edit]

The CPU voltage and speed is stock (so 1.35, I suppose)

I was under the impression that most DDR2-800 is being sold with a vendor-overclock, like how firms used to sell a shop-guaranteed PIII or Slot A Athlon guaranteed to work at a specific overclock and voltage.

Out of 36 2x2Gb DDR2-800 kits being sold on Newegg right now, I note only 12 are listed with a voltage of "1.8v", "1.8-1.9", "1.8-2.0", or "1.8-2.2". Two thirds of the products on the market won't even pay LIPSERVICE to the standard that DDR2 is 1.8v!

I hate to imagine people buying Dells or HP boxes that lack the voltage knobs necessary to make that two-thirds of the market compatible.