Normally when a miner starts, gpu boost 3.0 boosts the core clock to 2025-2075mhz and when miner quits it comes down to original state, but while cooldown the gpu gets boosted even more for a split second causing the nvidia driver to crash. the problem here is that when start or quit, there is not enough power ( i mean software/driver limit). All gpu follow a base profile ( _GLOBAL_DRIVER_PROFILE (Base Profile) ) here there is a key that states ' CUDA - Force P2 State' its ON by default. you just need to change the state to OFF.

You can check the power state by command-line to get current state