search

damonge / CoLoRe

CoLoRe - Cosmological Lofty Realizations
GNU General Public License v3.0
17 stars 13 forks source link

Different sims obtained when using different number of MPI nodes #28

Open andreufont opened 6 years ago

andreufont commented 6 years ago

I get different simulations when I use the same seed, but different number of MPI nodes.

Is this a bug, or a well known feature?

I found this while investigating issue #25 on quasars with Dec=NaN, since I got different number of bad Dec when using MPI and not using it.

damonge commented 6 years ago

@andreufont Yes, this is to be expected. The seeding happens differently depending on the number of nodes. Is this a problem?

andreufont commented 6 years ago

@damonge - no, not really. In general is nice when you get the same answer regardless of OpenMP / MPI setting, but as long as it is documented, this should not be an issue. I just though it was a potential bug that could be related to some of the other issues, but we can close this issue.

damonge commented 6 years ago

OK, thanks. I'll leave all issues open for the time being until we find the reason for the NaNs.

andreufont commented 6 years ago

Should we also close this one?

damonge commented 6 years ago

Let's leave it open in case someone else encounters this "issue"

Laura-Casas commented 7 months ago

@cramirezpe - mira lo que me ha enseñado @andreufont

cramirezpe commented 7 months ago

@Laura-Casas, thanks for reminding this... I was aware but at some point, I forgot it.