GPU utilization for CUDA

how to get maximum of percent of multiprocessors or kernels used by GPU at any point of time? nvidia-smi gives percent of time one or more kernels are active but I am looking for percent of multiprocessors. how do i get this metric? what is the compute utilization in visual profiler?
Basically i have 2 apps, if I run them simultaneously it takes longer than running them individually. that means each app takes 100 ms for completion without any other app simultaneously running but running them concurrently takes 150 ms. So i want to see how many cores are used by each of them. Any other way to debug this issue?

From my experience nvidia-smi only runs at seconds interval which is too big.
I usually run it like this: watch -n.1 nvidia-smi
which gives better resolution. I also mostly watch the Watt and not just the utilization.

Other than that - use nvprof to better see what your app is doing.

Eyal