Parallel computing efficiency - Number of cores or MPI processes?

Question

I'm testing a parallel program on a 2 + 2 (hyper threading) cores machine. The parallelization is effected through MPI routines.

While evaluating the efficiency of the parallel code, do I have to divide the speed up by the number of cores or the speed up by the number of MPI processes?

Can anyone help me please?

Answer 1

I think correct answare is that p is always the number of processor/cores because the efficiency deacribes how your code's performances tend to the ideal values. So the genaral formula for the efficiency is

E=your_performance/best_performance

Best speed (ideal case) is the number of processors/cores p.