Sup doods, I've been tinkering with my 580SC for a while, and some questions arise.

First: Would it be best to keep your voltage at max and just get the best possible STABLE OC on it and leave it like that? My temps never reach anything above 65 on max load no matter what voltage I use, going from stock voltage to max allowed only bumps the temps 3 degrees or so.
 
Second: My stable OCs become unstable for no reason. For example, the other day I had it at 900 core(forgot shader #) 2150 memory, and at 1.113v. This OC ran vantage no problem and gave me no errors over 15 mins of stress testing , maxing out at 61c. This OC however I did with the windows open, so my room was kinda cold.
The next day I tried vantage again with the same OC and it would crash 2 seconds into the first test, doing a stress test I would get artifacts and the max temps would reach 66c or so.

Could the 5 degree difference in temps be enough to destabilize the whole thing? I know with OCing you're walking on a tight line, but is it really that sensitive? What exactly does increasing voltage do regarding stability? Is it just part of the magic temp/vcore mix that makes an overclock stable?

Right now I left it at 1.113v and 880core 2150 memory and ran vantage  a couple times this morning before work, came through fine, no errors. This was in a warm no-t-shirt-necessary temperature room though. The 900 core was in a warm-sweatshirt-required room with a cold breeze blowing down right on the PC.
 
Thanks for any ideas/suggestions/advice/discussion.

1. It's a risk, but I'm not sure if it is a small risk or large risk.  You can't measure the temperatures of the VRM which is what heats up as you increase voltages (as well as GPU core).  But the GTX 580 has a good heat sink that cools everything so you should be fine when running 1138 mV.
 
2. Yeah increase temperatures can destabilize overclocks.  During the summer my maximum GPU overclock for Vantage on my GTX 480 was 885/1770/1000.  Now, with the lower ambient temperatures, I can bench at 890/1780/1000.  15 minutes of stress testing is not nearly enough.  If you want to ensure a stable overclock, run the EVGA OC Scanner at full screen, enable artifact scanner, enable power draw and run it for at least an hour.  You might also want to try the Cryostasis Tech Demo.  It's an outstanding GPU stress test despite how short it is.  The insane amount PhysX calculations prove unstable overclocks all the time.  Back when I had a single GTX 260 CORE 216, I had an overclock that passed ATItool and OCCT but crashed during that game.
 
3DMark Vantage is not close to a valid stress test.  I use slightly unstable overclocks for benchmarks just to squeeze out more points, but for normal use I only use a stable overclock that passed the stability tests I mentioned above.

Oh I definitely ran EVGA OC Scan to overload the card, but sometimes I'd get decently stable short runs with the OC scanner, but 2 seconds into Vantage it would crash.

I'm not gonna go with a homebrew watercooling solution, but are there any decent closed-loop no maintenance systems for GPUs like the H50 cooler for CPUs from corsair?