I can say with certainty that most GPU stability testing tools only check memory indirectly- Like a CPU burn in with small data sets. With 2 or 3 GB of memory, the chances of an indirect hit at pretty low. You can see the results of this in games that have high graphics memory usage being far more sensitive to bad overclocks than tools with virtually no graphics memory footprint.
So does anyone know a method to test the memory stability directly with a simple graphical test? Hell it doesn't even have to be graphical.. Load the cores for relative temp and clocks then swap 2 GB with the ram a million times and check for changes.
(a directcompute benchmark with large data sets is probably the right answer but I'm not familiar with any)