Originally posted by: SystemAdmin

[rlinfati said:]

Hi

do you have a benchmark of cplex in 32 vs 64 bits?
or how i can do a (good) benchmark

and for windows vs linux ( but... i think i can not see diferences here )

thanks!
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

[notdarkyet said:]

My experience for the standalone CPLEX is: the 64 Bit executable for Unix (Linux and Solaris) yield a branching tree different from that of the 32 Bit Linux executable.
My only idea why this is the case: minuscule differences in flowating point arithmetic lead to different branching decisions. Of course I cannot verify this assumption.
With different branching done at one point in the tree naturally it depends on the problem which one is faster.

So benchmarking will be difficult - if not impossible.
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

[achterberg said:]

Notdarkyet,

your assessment is correct. In principle, there should be no performance difference between differen operating systems and pointer widths (apart from the Windows memory manager being inferior to the one of Linux). But due to small differences in floating point calculations the search tree can look completely different for the same model on different platforms and thus, the solving speed for an individual model can vary greatly.

There is not much you can do about it. In order to assess this effect by benchmarking, I recommend to run lots (say, 100 to 200) of different problem instances. Then, calculate the geometric mean of the run time. This geometric mean should be similar over all platforms.

#CPLEXOptimizers
#DecisionOptimization