concurrent different kernel blocks & HyperQ with old CC3.0 OpenCV build

Hi,I have 3 questions.

1.I changed my GPU from GTX 650 Ti Boost to GTX 750 Ti,I searched the driver and found both of them use the same version of driver(344.65),my OS is win8.1(32 bit),so do I need to re-install CUDA 6.5 and the driver if I change to 750 ti?or just plug & play?

2.I have OpenCV 3.0 dlls or libs in which cu files compiled with flag sm_30(without sm_x,x>=35) are incorprated ,I also have cu files compiled with flag sm_30 of my own ,do I need to rebuild these files when I change my hardware from CC3.0 to CC 5.0 in order to enable HyperQ ?cuz rebuilding is very time-consuming and I don’t want to do it again.

3.Can thread blocks from different kernels run concurrently on one SM?or they can only run concurrently on different SMs?