I'm running 4x 5830 @ 950/300 but having a small problem with occasional crashes on the main GPU. Basically the graphics card driver crashes and reloads itself (it does this fine) but mining on that particular GPU will stop and not resume correctly.
Why does it crash? Is it because idle clock for memory is too low and windows does require the main GPU's memory to be "un-idle" sometimes?
Is there anyway to automatically recover from the crashed driver?
Lower the aggression on that card alone, it may be too hard for the card to mine and draw the desktop at the same time.
If this is in a tower config, try laying the case on the side -- the 'main' card may be soaking up heat from the cards below it and overheating.
Lower that cards' GPU/mem clocks and undo any voltage changes, the card may not be as stable at high clock/high heat/low mem as the others.
Swap the crashing card to a different slot, see if it still crashes. It may be the slot itself.
Above all check heat and rejection rate; a higher than average rejection rate often means a card is overclocked too high or somehow defective.
... as for making the miner restart when the driver crashes, try using of the scripts floating around the forums that will kill and restart your miners ever XX minutes.