how to see Multiprocessor Utilization all time?

when I use the cuda profiler to get the Multiprocessor Utilization, sometimes I can find it in kernel latency, but sometimes I can not find it. How to make the cuda profiler show it all time?