i am currently running a fairly big transient problem, initially on windows (v3.150a) and the simulation is progressing well if a little slow (courant numbers, piso correctors etc etc all look good) due to the time aspect i decided to run it on a far more powerfull linux machine but for some reason the results i am getting are totally different even though the model is exactly the same, any ideas?

Are you running the same patch version of STAR on both systems? You may have a bug correction in one that has not been added to the other.

Maybe the REAL (or DOUBLE) values are represented with different precisions on the two platforms? This would lead to different solutions if rounding-errors play a role.

With any platform (Windows, Linux, whatever) the user can decide to work in single precision or double. Single precision on Windows is the same as single on any other platform and same goes for double. Unless you actually picked single precision on one and double on the other, then that's not why the answers are different.