| posts: 831 | Location: Greece
I never found an NVIDIA card that can do below 10% GPU overclock at stock voltage. For the memory clock thought, I've seen everything starting from only 1-2% to as high as 25%.
Anything that is REALLY stable and has the temperature under control (including PCB and power circuit temps) should be healhty for the card.
Personally this is how I work: Start at about 8% GPU overclock with a temperature monitoring window (usually GPU-Z) and start a stress testing utility (usually ATItool). Every 2 minutes I start increasing the overclock 1 step, until I reach the GPU frequency that hangs/artifacts withing this 2-minute window. When I detect this, I immediatly stop the stress test and downclock the card to the previous value and I now start decreasing the overclock step-by-step until the value it completes the 1-hour stress test (usually it is only 1-3 steps below the previous frequency). If the card is for my mashine I select the maximum stable I find, but if I build it for someone else, I downclock it 2 more steps for more safety (actually the only problem is that if that frequency is not 100% stable 24/7, then whenever you detect some artifacts you must immediately stop gaming and decrease the overclock one step. If you just keep watching at them, they'll probably damage the card).