Getting different results when running on Windows and Linux

[stroop]

I am simulating electromagnetic waves through a waveguide that employs the VectorHelmHoltz solver and I am getting different results when I run it on Windows versus when its run on Linux. I have the latest version of both but these versions are different, that is the windows version is older but the change is ever so slight (the imaginary part of the magnetic boundary load is taken into account, that is it was changed from 'Magnetic Boundary Load im 2' to 'Magnetic Boundary Load im 2' so the extra space between Load and im was removed). So I don't see why there should be such a significant different in the values especially since my case doesn't even change or use the value of the magnetic boundary load im in it's initial conditions.

I have attached my case, the output and the data file from both systems below.

As you can see most values are just slightly different but not for the 'integral of div poynting vector' why is this?

The rest of the files are posted in the following post.

[stroop]

Here are the remaining files

[raback]

Hi

You could try with nightly build for Windows at:
http://www.nic.funet.fi/pub/sci/physics ... n/windows/

-Peter

[stroop]

Thanks for your suggestion, I tried the following nigtly build: elmerfem-nightly_Windows-AMD64.zip
but it still doesn't match the output given by the Linux version and is the same as the old windows version.
I have attached the output and the data file produced by the new nightly build.

[mzenker]

Hi,

as far as I see, at least the first five digits of the numerical values in the .dat files match, except for the 5th one, which is off by a factor of two, but which is very small (1e-22). So IMHO you cannot say that there is a significant difference, unless the difference for the 5th value is your main problem.
BTW do you use the exact same mesh file for both simulations?

HTH,

Matthias

[raback]

Hi

Note also that Windows and Linux binaries are not doing bitwise the same thing. The order of some operations may vary etc. which may affect the results. This could well affect the 5th digit particularly if you're working with extreme range.

-Peter

[stroop]

I thought any difference in the number was indication of a problem but since you have confirmed that it is ok I guess it's not a problem any more - Thank you for your help!