How do you run Potjans_2014 in full scale?

[bjourne]

How do you run Potjans_2014 in full scale?

[heplesser]

@bjourne Did you change local_num_threads in sim_params.py? With the default value of four and 300 million connections in total, you should not hit the limit at 134 million per virtual process (thread), unless you try to build the model once in a single Python instance without cleaning up with nest.ResetKernel() in between. You could try to increase local_num_threads.

If that does not help: How did you install/compile NEST?

BTW, for questions like these our NEST User Mailing List is the better place (https://www.nest-simulator.org/community/).

[bjourne]

Thanks. I changed the number of threads to 8 by setting "local_num_threads": 8,. This however seem to affect the simulation and I don't get the same results as when I run it with 1 thread.

[heplesser]

Due to the way random number generators work in NEST, different numbers of virtual processes (MPI ranks times OpenMP threads) will yield different random number sequences and thus different detailed spike patterns. But if you look at measures such as average firing rates, inter-spike interval distributions or cross-correlations, they should not differ significantly for different numbers of parallel processes.