Fro a specific user's model with only 2400 node points took 43 min to run the analysis:
Load cases analyzed ................... GR T1 T2 W1 W2 W3 W4 P1 P2 SN
This seems a little excessive for this number of node points. Note, this model combines a STAAD file into AutoPIPE using PipeLINK.
After reviewing this specific model, AutoPIPE does take a considerable amount of time to perform an analysis. After testing, the multitude number of beams in this model have a much higher effect on the analysis time than the piping nodes.
AutoPIPE is simply requiring a lot of cycles for stiffness decomposition. This exact number of cycles or lines of code required to iteration of a load case for non-linear is in the neighborhood of tens of billions of lines of code per iteration of any one load case. Therefore, the calculations for a model with both pipes and beams is very CPU intensive, which would explain why the analysis take so long.
Another way of saying this; the stiffness matrix reduction driver (subroutine factor) is what takes the most time to compute, and it is iterated many times during analysis of a particular model.
However, every model is different, if you have a model that is taken a considerable amount of time for no apparent reason, log a Bentley Service Request, included an AutoPIPE APC file, and mention this WIKI page
One possible solution would be to increase the force and or displacement tolerance in the nonlinear analysis dialog in an effort to reduce the number of iterations required for each load case.