.. _v3.8_benchmark_results:

NEST 3.8 performance benchmarks
================================

.. important::

   NEST version 3.8 benchmarks of the Multi-area-model and HPC benchmark model have been updated due to errors in the analysis! Please see below for corrected versions.

NEST performance is continuously monitored and improved across various network sizes.
Here we show benchmarking results for NEST version 3.8 on Jureca-DC [1]_.
The benchmarking framework and the structure of the graphs is described in [2]_.
For details on `State Propagation` (i.e., `Simulation Run`), see the guides :ref:`built_in_timers` and :ref:`run_simulations`

Strong scaling experiment of the Microcircuit model [3]_
---------------------------------------------------------

.. grid:: 1 1 1 1

   .. grid-item::
       :columns: 10
       :class: sd-align-major-center

       .. image:: /_static/img/mc_benchmark_NEST-v3.8.png

.. grid:: 1 1 1 1

   .. grid-item::
       :columns: 10
       :class: sd-align-minor-center


       * The model has ~80 000 neurons and ~300 million synapses, minimal delay 0.1 ms
       * 2 MPI processes per node, 64 threads per MPI process
       * Increasing number of computing resources decreases simulation time
       * Data averaged over 3 runs with different seeds, error bars indicate standard deviation
       * The model runs faster than real time [4]_




Strong scaling experiment of the Multi-area-model [5]_
-------------------------------------------------------

.. grid:: 1 1 1 1

 .. grid-item::
       :class: sd-align-major-center
       :columns: 10

       Dynamical regime: Ground state

       .. image:: /_static/img/mam_ground-state_benchmark_NEST-v3.8.png

       Dynamical regime: Metastable state

       .. image:: /_static/img/mam_metastable-state_benchmark_NEST-v3.8.png


.. grid:: 1 1 1 1

 .. grid-item::
       :columns: 10
       :class: sd-align-minor-center

       * The model has ~4.1 million neurons and ~24 billion synapses, minimal delay 0.1 ms
       * It can be run in two different dynamical regimes: the ground state and the metastable state [5]_.
       * 2 MPI processes per node, 64 threads per MPI process
       * Steady decrease of run time with additional compute resources
       * Data averaged over 3 runs with different seeds, error bars indicate standard deviation




Weak scaling experiment of the HPC benchmark model [6]_
--------------------------------------------------------

.. grid:: 1 1 1 1

   .. grid-item::
       :columns: 10
       :class: sd-align-major-center

       .. image:: /_static/img/hpc_benchmark_NEST-v3.8.png


.. grid:: 1 1 1 1

   .. grid-item::
       :columns: 10
       :class: sd-align-minor-center


       * The size of the network scales proportionally with the computational resources used
       * Largest network size in this diagram: ~5.8 million neurons and ~65 billion synapses, minimal delay 1.5 ms
       * 2 MPI processes per node, 64 threads per MPI process
       * The figure shows that NEST can handle massive networks and simulate them efficiently
       * Data averaged over 3 runs with different seeds, error bars indicate standard deviation


.. seealso::

   * Guide to :ref:`Built-in timers <built_in_timers>` and :ref:`run_simulations`.

   Example networks:

   * :doc:`Microcircuit Model <pd14:index>`
   * `Multi-area model <https://inm-6.github.io/multi-area-model/>`_
   * :doc:`HPC benchmark <nest:auto_examples/hpc_benchmark>`

References
----------

.. [1]  Juelich Supercomputing Centre. 2021.  JURECA: Data Centric and Booster Modules implementing the Modular
        Supercomputing Architecture at Jülich Supercomputing Centre. Journal of large-scale research facilities,
        7, A182. DOI: http://dx.doi.org/10.17815/jlsrf-7-182


.. [2]  Albers J, Pronold J, Kurth AC, Vennemo SB, Haghighi Mood K, Patronis A, Terhorst D, Jordan J, Kunkel S,
        Tetzlaff T, Diesmann M and Senk J (2022). A Modular Workflow for Performance Benchmarking of Neuronal Network Simulations.
        Frontiers in Neuroinformatics(16):837549. https://doi.org/10.3389/fninf.2022.837549

.. [3]  Potjans TC. and Diesmann M. 2014. The cell-type specific cortical
        microcircuit: relating structure and activity in a full-scale spiking
        network model. Cerebral Cortex. 24(3):785–806. DOI: `10.1093/cercor/bhs358 <https://doi.org/10.1093/cercor/bhs358>`__.

.. [4]  Kurth AC, Senk J, Terhorst D, Finnerty J, Diesmann M. 2022. Sub-realtime simulation of a neuronal network of natural density.
        Neuromorphic computing and engineering 2(2), 021001
        https://iopscience.iop.org/article/10.1088/2634-4386/ac55fc/meta

.. [5]  Schmidt M, Bakker R, Hilgetag CC, Diesmann M and van Albada SJ. 2018. Multi-scale
        account of the network structure of macaque visual cortex. Brain Structure
        and Function. 223: 1409 https://doi.org/10.1007/s00429-017-1554-4

.. [6]  Jordan J, Ippen T, Helias M, Kitayama I, Sato M, Igarashi J, Diesmann M, Kunkel S. 2018.
        Extremely scalable spiking neuronal network simulation code: From laptops to exascale computers.
        Frontiers in Neuroinformatics. 12. https://www.frontiersin.org/journals/neuroinformatics/articles/10.3389/fninf.2018.00002