I have attached a spreadsheet which is a useful tool for estimating simulation run time, this calculates the time-step (typically based on the smallest sphere in the simulation) and the user can input the time per iteration per particle along with the required simulation run time and estimated number of particles.  The time per iteration is the time it takes for EDEM to complete 1 time-step calculation and this tends to be a constant value for any given hardware on a 'per particle' basis.

The time per iteration per particle is simulation and hardware dependent, typically this would be between 0.5-1 microseconds for a 12 CPU simulation and a GPU can be 10x faster than CPU.

The above graph shows that for the test case of varying particle numbers the time per iteration per particle is constant above 1e4 particles, below this the efficiently of using GPU drops away.