Dear Stephen,

That is true, yeap, I've seen the paper earlier, but it was done only for the Freesurfer, CIVET, FSL.

So, there was another one claiming that "to avoid such effects, software developers recommend using the latest release version of the toolbox and do not to mix versions, platforms, and OS versions within one study."
But I found nothing about SPM and all related add-ins (as CONN) to be aware of.

Also, I should say, that I've checked several times and CONN results do reproduce within one machine in time, however
bias depending on the processor core.

And, as CONN is the Matlab toolbox, and it is GUI - it is a huge problem to run it on the cluster. Instead of it on the cluster,
one can run Freesurfer, Nilearn, FSL, etc. Thus, for my perspective, CONN is mostly for the local machines; but in large datasets, if you spread the computation among your colleagues - you are subsequently facing the results bias.
Therefore, it is uber important to keep that in mind, especially in toolboxes providing GUI.


Anyways, will hope that this problem will be solved and thank you so much for the response!



warm regards, Kate