NVIDIA P$100$ GPU has $3584$ single$-$precision floating point cores with a peak $10\mathrm{.}6$

teraflops of single$-$precision floating point computation $(1$ teraflop $=1000$ gigaflops $=1012$ floating

point operations per second$).$ The interface to global memory has a bandwidth of $720$ GBytes$/$sec$.$

How many floating point operations must the GP$100$ perform on each float loaded from global

memory if it is to achieve its peak floating point operation rate $($and not by limited by memory

bandwidth$)?$