Failed OC now unstable at stock

Please forgive me if this is an inappropriate forum…

Let me begin by apologizing for the length of the following narrative and begging any readers to bear with me. I’ve been trying to avoid throwing myself on the mercy of the forums for a couple of weeks with no luck and I’m approaching the limits to RMA if I need to. If you’re willing to help but don’t want to read the whole story skip to the second set of system spec’s.

That said, here goes:
A friend and I built my rig about 3 years ago and apart from a small SNAFU, in which my bios decided to under-volt my RAM which lead to a slow and silent corruption of my windows install, Grond (my rig) kept up with the games I was playing very well. Until I hit a performance wall with Witcher 2 a moth ago (auto-detect said “low” but it was totally playable at “medium” settings).

Grond as stable for ~3 years:
Antec Solo Mid Tower Case (and appropriate case fans)
Corsair HX 1000 W PSU
EVGA nVidia 790i SLI mobo
Intel Core 2 Duo E8600 Wolfdale 3.33 GHz
Arctic Cooling Freezer 7 Pro CPU Cooler
4 Gig (4x1Gig) OCZ DDR3 2000 (PC3 16000) Dual channel RAM
EVGA GeForce GTX 280 1G 512-bit GPU
(HDD and DVD+R as well)
Running Vista 64bit Home Premium

Everything was on stock or auto settings except the RAM which needed 1.8V and the BIOS wanted to set for 1.65V

After hitting the wall I started working on upgrading my RAM and adding another GTX 280 as pre-planned when we built the rig. Of course I couldn’t find a new GTX 280 and had to be satisfied with a refurb which is why I’d like to get this resolved fast enough to RMA either the RAM or GPU if I need to.
Before I did anything I ran the “standard” stability tests I found prescribed on the forums i.e. Memtest86+ (default), Prime95 (Long FFT in place) and 3Dmark Vantage with RealTemp dumping to Excel. Memtest and Prime both ran for ~24 h without faults (Prime running 2 workers 1 on each “CPU” (core?) and 3Dmark ran 3 full test loops (P 13189, 10957 Graphics, 33927 CPU [PPU not disabled]; P9466, 10899 Graphics, 6788 CPU [PPU disabled]) without errors. After that, as prescribed on forums, I updated my drivers and BIOS (I don’t remember what version I had but I successfully flashed it to SZ17) and stability tested again before installing any hardware (with similar results to above).
That done I installed the new RAM, G.Skill DDR3-2000, PC3-16000, 8 Gigs (4x2Gig). I checked my BIOS and set the manufacturer recommended voltage of 1.65V and left everything else alone. I then retested with Memtest for ~24 hours with no errors. Last but not least I installed the “new” GTX 280 into PCIe slot 2 (the bottom slot on my mobo) and hooked up power and the SLI bridge for 2 way (slot closest to the back of the case). When I booted back up I got a BSoD with something about a display driver problem and a memory dump but it rebooted fine and nVidia control panel (with ntune which I got with the above mentioned driver updates) allowed me to enable SLI as expected. I ran 3Dmark again and saw great frame rate improvement with the graphics tests and better benchmarks (P131177, 191134 Graphics, 6814 CPU [with PPU disabled]).
Excited, I ran the Witcher 2 autodetect again which still suggested “low” settings though it ran pretty smoothly on “high”. This is where I got greedy, I’d noticed that RealTemp was showing core temperatures of low ~40C during game play, and 55C during Prime tests and I’d seen a lot of people running my e8600 at 4GHz without voltage changes… So I started reading about over-clocking my CPU a bit hoping to turn off the GPU dependant PhysX and put that load on my processor to free up the GPU’s. Ultimately hoping to get another couple of years out of my rig… I found some people who’d had good luck using ntune/nVidia control panel to achieve FSB over-clocking so I gave it a try. I just bumped the slider up enough to hit 3.4 GHz on the processer 340 (x10) and hit “apply”.
Unsurprisingly, my system crashed. On reboot I checked the BIOS to make sure everything had gone back to normal, then tried to boot into windows. I ended up having to use my Vista DVD to repair a corrupted registry but finally got my computer back up and running. I figured, “Lesson learned” and gave up on over-clocking and fired up The Witcher 2 again. Less than 15 minutes later the game just vanished, no freezing, no lockup, no windows warning, no BSoD just playing the game one second and looking at my desktop and the launcher the next.
So I started running Pime95 (as above, large FFT’s in place), both workers failed in the first 5 minutes. Then I ran Memtest, which ran fine (default tests) for ~24 h. I’ve made some progress since but I’m now totally stuck, see below:

Grond (my system) as it stands, unstable.
Antec Solo Mid Tower Case (and appropriate case fans)
Corsair HX 1000 W PSU
EVGA nVidia 790i SLI mobo (SZ17 BIOS) and 15.58 WHQL
Intel Core 2 Duo E8600 Wolfdale 3.33 GHz
Arctic Cooling Freezer 7 Pro CPU Cooler
G.Skill DDR3-2000, PC3-16000, 8 Gigs (4x2Gig)
EVGA GeForce GTX 280 1G 512-bit (x2, SLI) GPUs (Driver 8.17.12.9551)
(HDD and DVD+R as well)
Running Vista 64bit Home Premium (all updates)

Everything is running at stock/auto other than my ram which I’ve set for the manufacturers recommended 1.65V and my CPU core voltage which I set for the recommended 1.25V just incase “auto” wasn’t working for some reason. See BIOS Pics if I can get them posted.





Diagnostics:
Memtest86+ (v 4.2): Stable ~24 h
Prime95 (v 27.2): Large FFT in place, fail on at least one core (2 workers) anywhere from 13 min-5 hours. Failed in 5 minutes while still running ntune, so I uninstalled nVidia control panel, ccleaner’d installed the newest control panel (without ntune). Runs longer now, and even longer in safemode-command prompt but still fails (I’ve had one instance of coming back to check and finding that the computer had rebooted). Tried both Blend and Small FFT to distinguish between RAM, Processor and mobo but both small FFT and Blend return “No FFT lengths available in the range specified” (I’m still researching how to get around this).
3Dmark Vantage: Haven’t even tried since things went sideways
RealTemp: CPU’s 35C idle, 55C @ 100% load; GPU 85C during game play 50C idle
EVGA Precision (v 2.1.2): This is new to me, only thing I’ve noticed that stands out is that GPU 2’s memory usage is maxed anytime I have SLI enabled and 0 when disabled, no in between. SLI enable/disable seems to have no bearing on Prime failure though.
EVGA On-Screen Display server (v 4.3.1): Which I assume will be useful if I ever decide to risk OC’ing either my GPU’s or CPU again.

Top priority is to get everything stable again, though I’d still like to figure out how to get another year or so out of this rig.

Any help or input would be appreciated,
NOT being very Zen atm!