I RMAed one of my EVGA 780Ti SC cards recently because it was restarting under heavy load (such as 3dMark Fire Strike Extreme and Ultra). I had 2 of them in SLI so I tested each card individually and found that it was only occurring on the one card. I used the same slot when testing both cards as well as the same power cable to try and make the test as apples to apples as I could. I also tested with AIDA64 and it would restart and crash the test in under 2 mins with the suspected card and not at all in over 40 mins with the other. So, two weeks go by and I just got my replacement card in the mail. I get it installed and do some testing as I know they are re-certified replacement cards and not new cards. Turns out that the replacement card is crashing and restarting as well during the AIDA64 test, but this time I can loop Fire Strike test 2 for over and hour without issue. Should I be concerned with the AIDA64 test?
 
I have not touched any settings on the cards, other than fan speed control, but they are factory OCed from EVGA.
During the original testing I was on 352.86, but tried 353.30 and then ended up on 353.38 before RMAing. I am currently still on those .38 drivers.
My CPU is overclocked, but I returned to default settings when I was testing.
 
By restarting I mean that the screen goes black, comes back on with the driver crash message, and I can hear the fans go silent and then return to full speed after the screen comes back on.
 
System:
i7 4770k @4.5Ghz
2x EVGA 780Ti Superclocked
2x 8GB Corsair Dominator Platinum 1866Mhz
Samsung 840 EVO 500GB
2x WD Black 1TB
Kraken x60 Water Cooling
Antec HCG 900W

Thanks for your reply. AIDA64 is a full benchmark suite, but you can choose which items to test. I was hoping someone might have had experience with AIDA64 GPU stress testing before and had some insight into if it tests beyond the cards "margin of error" if you will. I know any overclock can be or can become unstable and since these cards are overclocked from EVGA, some might be more "dialed in" than others. I do not know how they test the cards when they overclock them to assure they are stable. Maybe AIDA64 is extreme for this purpose?
 
I did play The Witcher 3 today on maxed settings and left the game running when I took breaks and did other things. I imagine it was running maybe 5-6 hours straight. I will take your advice as well and run 3Dmark for some extended loops to see how that goes as well.
 
The only part that nags me about all of this is that my other cards seems to have no problem at all with the AIDA64 GPU test. I just don't know if that means the new card is "on the edge" of going out, or if it simply doesn't have as tight of an overclock on it.

Ya I think I am going to. Sucks it will mean another 20 dollars in shipping and 2 weeks without gaming. Makes me rethink the whole factory overclock cards in the future.

Just a follow up to say that EVGA is taking care of me. I didn't need to complain either. I simply asked if I would have to pay the shipping again and was informed that I would be upgraded to the advanced RMA method for free. Fingers crossed this next card will be better.

2nd replacement card is doing the same thing. Help me out here guys. What am I missing / not thinking of? I am at a loss now as to what to think. The only thing for sure is that these replacement cards and my original card are not created equal. 2 mins is all it will stress test before the card resets. AIDA64 too hard of a test for GPUs? I haven't been able to find any other complaints about that. In fact, I have seen the opposite where people have said it passes AIDA64, but crashes in games. All I know is my original card passes with flying colors and each of the replacement cards I have recieved fail in about 2 mins.

I think I just confirmed my suspicion. I downclocked the core to the base core speed (according to the specs from EVGA) using precision X. It then ran AIDA64 fine for over a half hour. I think the overclock on these are just not done well or are degrading. I am not sure I will buy another factory overclock if this is the headache it can produce.

I run with the fan curve a bit more aggresive than stock. So, the max temp I was seeing in AIDA64 during stress was about 72c.