I am running a Fortran code on HPC machines with a lot of cores (say 36). Using parallel MKL, I set the number of cores to be used with the environment variable MKL_NUM_THREADS, while the OMP_NUM_THREADS variable is set to 1. Despite that, my code sometimes runs only with 1 core. Do the MKL adjust the number of cores themselves based on free memory left? Also is their memory scaling with the number of cores kinda linear or not? Do you think that the problem is something else?