Dear Ardavan Oskooi. Thank you very much for your answer. I tried setting split_chunks_evenly=False but it did not improve too much the RAM usage.
Best regards, Juan. From: Ardavan Oskooi<mailto:ardavan.osk...@gmail.com> Sent: Thursday, November 26, 2020 6:18 PM To: Juan Ramón D<mailto:deop...@hotmail.com>; meep-discuss@ab-initio.mit.edu<mailto:meep-discuss@ab-initio.mit.edu> Subject: Re: [Meep-discuss] RAM issues To improve memory usage in your simulations, try setting "split_chunks_evenly=False" in the Simulation constructor which should result in chunks that are better matched with the gold layers. You can use the visualize_chunks<https://meep.readthedocs.io/en/latest/Python_User_Interface/#data-visualization> routine to compare the chunk layouts (equal vs. unequal). For details regarding the chunks framework used for paralell/MPI simulations, see Features/Parallel Meep<https://meep.readthedocs.io/en/latest/Parallel_Meep/> as well as Section 3 "Data-Driven Load Balancing" of https://arxiv.org/abs/2003.04287. Finally, note that the memory management of the chunks for Lorentzian susceptibility materials (i.e., gold) is currently suboptimal: https://github.com/NanoComp/meep/issues/1084. More memory is being used than is actually required by the simulation. We are hoping to fix this soon. On 11/26/2020 4:22, Juan Ramón D wrote: Is there any way I can reduce the RAM usage for such a simulation, to get a higher resolution? Could it be because of a different reason? I can only think of reducing the cell (PML layers) and using non-dispersive models for the materials, but both methods compromise the results of my simulation. -->
_______________________________________________ meep-discuss mailing list meep-discuss@ab-initio.mit.edu http://ab-initio.mit.edu/cgi-bin/mailman/listinfo/meep-discuss
