Different results with same code and data


I am worried about different results (very different actually) given by same code and same data on two slightly different platforms.

  1. one is 32 bit Intel processor with Fedora 9 with gcc 4.3.0, root 5.25/02,

  2. another is AMD opteron 64 bit processor but 32 bit Fedora 8 with gcc 4.1.2, root 5.22/00.

I fitted a histogram with gaus+pol3 and I got quit different results: in first I got 6% fit probability and in second I got 74% fit probabilty. Other fit parameters are also very different.

Some how MINUIT is going through more steps (as shown by output) in second case than in first.

Now, what could be the reason and which one should be trusted?


Could you post a short but RUNNING script (and the data file) reproducing your problem?