A metric from Nsight Compute, Waves Per SM is an average of the number of waves of all SMs.
Here is my logic in comprehending this as a metric for load balancing:
If all SMs have the same number of waves (perfectly balanced load), Waves Per SM would be an integer.
Is it only when certain SMs have additional number waves where the average ‘Waves Per SM’ gets decimal values.
Following this logic, a smaller decimal value would indicate that fewer SMs have that additional wave.
Thus am I correct in understanding that, smaller the decimal value of ‘Waves Per SM’, the worse the load balance is?