Originally posted by: billtheguy

For the same model, same settings, and same machine same condition, CPLEX 12.3 is much slower than 12.1 (see attached figure). Anyone has a clue?

I also tested the same model using Gurobi4.5, Gurobi is even faster (<100seconds). Is CPLEX falling behind?
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

This is typical for MIP. A new version may only apply slightly different decisions at some point, and the whole search tree evolves differently, sometimes with good and sometimes with bad luck.
You cannot evaluate the speed of a solver (or of some version of a solver) with only a single model. Performance variability of MIP models is just way too large. You need to evaluate hundreds or better thousands of models to really find out whether CPLEX 12.1 or CPLEX 12.3 is faster on your models.

If you observe such a degradation consistently on a number of models, then you may want to play around with parameter settings or the auto tuning tool that CPLEX offers.

Additionally, it would be nice if I could take a look at your model. Maybe I can see something interesting in there. If possible, please send it to achterberg at de dot ibm dot com.
Thanks,

Tobais
#CPLEXOptimizers
#DecisionOptimization