Different results for different runs

[Ken_Ken]

I try to use Intel Fortran XE 2011 Composer  with MKL on windows 7.When I do the optimization with my code I achieve different results for different runs with the same initial values. I don't have this problem when I run Compaq Visual Fortran with IMSL on windows XP. Does anybody have an idea what is this? Is this a floating point problem? Can I solve it? and how?

Thank you in advance

[Steven_L_Intel1]

I can think of two possible causes of this.  1) You have uninitialized variables that are picking up garbage. 2) Because of Windows' Address Space Layout Radomization, allocated storage may be at different alignments on different runs. This affects vectorization and can change results slightly.  In MKL 11 (coming out with the "2013" compiler in a couple of weeks), there's a new option for improved consistency of results that may help with the second issue.

Can you determine where in the program the results start to diverge?

[bmchenry]

Need for you to give us an idea of what you mean by 'different results'?
large or small?
What level of optimization? (and what level on the ole Compaq?)
and as Steve mentioned, when do you find the results begin to change?
Meaning is it a small difference which accumulates into a larger issue?
Or is it a single large step/initialization type issue?