Does XGBoost produce the same results if I use different number of cores?

Question

I've installed the same exact version of XGBoost (0.4) on two machines. The only difference between the two machines is the RAM and number of cores (8 vs 16). Using the exact same data, I cannot reproduce the same results. They are slightly different (fourth, fifth decimal). The seed is left to the default.

Answer 1

It is highly implementation specific but in general randomized algorithms, ran in parallel may behave differently when working with different number of cores (unless one forces synchronization of random number generators, which would slow down the process). So it is something that one should expect - the same applies to Random Forest model etc.