Reproducibility of simulations

[Bernhard Kaplan]

Dear MCX-users,

I have a question concerning the reproducibility of simulations.
I observed that running the same simulation (with identical parameters) leads to different results (measured as the sum of all output voxels, no matter if  --outputtype is fluence, flux or energy).
I of course leave the "Session" : "RNGSeed" value constant (btw does --seed overwrite this parameter or is it a different seed parameter?).
I also use the default -R -2 setting to avoid non-atomic operations (to eliminate sources of randomness), but I wonder if this option is active or is it only available on particular graphics cards and if mine (geforce gtx 980 ti) supports the non-atomic write feature (or is it a compilation flag I've missed?).
Obviously reproducibility is crucial for all sorts of scientific questions, so I'd be happy to understand how to get the exact same results for two runs of MCX.

I'm using MCX Version: 0.9.7-2 (Dark Matter - Alpha update 2) downloaded from github ~a month ago.

Many thanks in advance and best regards,
Bernhard