Dear colleagues,
We are procuring a new small computing cluster on which we want to primarily run NEST simulations.
Our models are medium in size (100s of thousands of neurons) but we run them for relatively long biological times (100s of seconds), so the network setup duration is mostly irrelevant for us.
What we would like to ask is if there is any insight into the current NEST simulator in terms of what are the hardware bottlenecks of these types of simulations. Are we going to do better with investing into more cores or higher clock speeds or more cache, or faster RAM? I understand that this is a complex multi-factor optimization question, but any insights or links to existing material that would allow us to form an informed decision would be greatly appreciated.
Best regards,
Remy Cagnol.
Dear Remy,
we have been assessing the performance of NEST for models of approximately the size you are mentioning, see for example herehttps://iopscience.iop.org/article/10.1088/2634-4386/ac55fc/meta and herehttps://www.frontiersin.org/articles/10.3389/fninf.2022.837549/full. Generally, NEST does not profit so much from very high clock speeds. As the dynamics of neuronal networks is a microscopically parallel problem NEST can exploit parallel hardware: A high number of cores with a big cache seems to be helpful getting very good performance. And indeed, internally conducted benchmarks indicate that even larger cache sizes (in comparison to the nodes used in the cited paper) can still yield improvements in performance. Additionally, when simulating larger networks the speed of data transmission between nodes becomes important for the performance.
Best
Anno
On 22.11.23 16:50, cagnol@ksvi.mff.cuni.czmailto:cagnol@ksvi.mff.cuni.cz wrote:
Dear colleagues,
We are procuring a new small computing cluster on which we want to primarily run NEST simulations.
Our models are medium in size (100s of thousands of neurons) but we run them for relatively long biological times (100s of seconds), so the network setup duration is mostly irrelevant for us.
What we would like to ask is if there is any insight into the current NEST simulator in terms of what are the hardware bottlenecks of these types of simulations. Are we going to do better with investing into more cores or higher clock speeds or more cache, or faster RAM? I understand that this is a complex multi-factor optimization question, but any insights or links to existing material that would allow us to form an informed decision would be greatly appreciated.
Best regards,
Remy Cagnol. _______________________________________________ NEST Users mailing list -- users@nest-simulator.orgmailto:users@nest-simulator.org To unsubscribe send an email to users-leave@nest-simulator.orgmailto:users-leave@nest-simulator.org
------------------------------------------------------------------------------------------------ ------------------------------------------------------------------------------------------------ Forschungszentrum Jülich GmbH 52425 Jülich Sitz der Gesellschaft: Jülich Eingetragen im Handelsregister des Amtsgerichts Düren Nr. HR B 3498 Vorsitzender des Aufsichtsrats: MinDir Stefan Müller Geschäftsführung: Prof. Dr. Astrid Lambrecht (Vorsitzende), Karsten Beneke (stellv. Vorsitzender), Dr. Ir. Pieter Jansens ------------------------------------------------------------------------------------------------ ------------------------------------------------------------------------------------------------
-
Anno Kurth -
cagnol＠ksvi.mff.cuni.cz