the relationship between the bandwidth and utilization on shared memory

There are some analyse results via Nsight,and i find that the value of the bandwidth and the utilization are not linear, I don’t understand the connection.for example:
program 1: shared memory bandwidth:5,111 GB/s-----Utilization:8(High)(assume there are ten degrees to describe the utilization )
program 2: shared memory bandwidth: 4,838GB/s----Utilization:7
program 3: shared memory bandwidth: 5,529GB/s----Utilization:8(High)
program 4: shared memory bandwidth: 5,038GB/s----Utilization:9(High)
program 5: shared memory bandwidth: 4,113GB/s----Utilization:6(Medium)
program 6: shared memory bandwidth: 4,089GB/s----Utilization:6(Medium)

So, any ideas please?

It appears you are using Nsight Eclipse Edition or NVIDIA Visual Profiler.

The CUPTI metrics for utilization have 10 linear buckets. The metrics report in bytes/cycle; however, you are trying to compare to bytes / sec. If the GPC clock changes then the scale will also change.