Part of your problem is the SSC's voltage of 1.12 or 1.18 -- see my rant below. It's not the fan or cooling, these chips are being clocked too high, period.
Try "OCCT 3.0" GPU test with error check enabled. I'm still testing and will post results later, but it's not pretty. I could recommend sticking to stock 576/1242/999 for reliability until you test it with OCCT. I recommend an overnight 8 hour test, and at least one hour for minimal testing.
For instance, it shipped with the memory at 1152. It appears that the last error-free setting is 1080, and one NEVER should run right next to the failure point, which means this card should run at 1053 maximally for stability. That's a full 99 MHz too much as sold.
I'm not even going to RMA -- what's the point. Clearly these SSC cards aren't qualified properly. nVidia set the clocks the way they did for a reason. $20 wasted over the normal card, oh well.
I'll rant too, about the voltage. Mine came as 1.18v -- and the cooler can't really keep up with the heat. The difference between 1.06v and 1.18v is about 80 (!) additional watts (measured 95 watt increase at the wall). No wonder people are getting PSOD (pink/purple screen of death). The SSC chips are cooking themselves. On top of that, I found very little overclock difference between 1.06/1.12/1.18 volts. A notch or maybe two at best -- a complete waste for 80 watts and frying cards. I flashed mine back to 1.06v permanently.
1.05v @ 675/1350/1044 for computing/rom defaults (step down from max settings). 684/1368/1053 for gaming (max settings no errors). Using OCCT gpu test, 576/1440
/999 freezes, 576/1404
/999 gets errors, and 576/1242/1080
gets errors. And they tried to shovel this as a 675/1458