For a similar model size (i.e., number of equations/DOF), the elapsed time for 2D model is much shorter than for 3D. I am using the same machine, same solver, and same memory/CPU configurations. Why is the solve time so much different?
The wavefront size plays a role in the solution time, since it indicates the equations that could not be eliminated by the solver during the solve process. The highest value of the wavefront is called the "maximum wavefront," and it directly affects memory requirements. The averaged, root-mean-square value is called the "RMS wavefront," and it directly affects solution time.
It should be noted that this only applies to direct solvers (EQSLV,SPARSE and EQSLV,FRONT).
As an example, the following models have the same shape and DOF:
(1) 2D plane183, 25k elements, ~75k nodes * 2 DOF/node = ~150k DOF
(2) 3D shell181, 25k elements, ~25k nodes * 6 DOF/node = ~150k DOF
(3) 3D solid185, 25k elements, ~50k nodes * 3 DOF/node = ~150k DOF
For a harmonic analysis of the three models listed above, here are the predicted values from RWFRNT (/RUNSTATS processor):
(1) max = 430, rms = 416, mean = 416
(2) max = 631, rms = 617, mean = 616 (47% larger than case 1)
(3) max = 634, rms = 618, mean = 617 (47% larger than case 1)
And here are the results from the solution time:
(1) CPU time = 224 s, wall time = 284 s
(2) CPU time= 324 s, wall time = 546 s (45% longer CPU time than case 1)
(3) CPU time = 336 s, wall time = 575 s (50% longer CPU time than case 1)
It may be helpful to imagine the model passing through a planar surface; at any point in time during this process, the wavefront indicates the number of degrees of freedom the solver has to contend with simultaneously.