Skip to main content
. 2022 Mar 1;15:785068. doi: 10.3389/fninf.2021.785068

Figure 3.

Figure 3

Contributions to the simulation time (sim time) for spiking neural network simulations with NEST (Section 3.1) where the number of MPI processes increases proportionally with the total number of neurons. Weak-scaling experiment running 2 MPI processes per compute node and 12 threads per MPI process, with a workload of 125, 000 neurons per MPI process (network model see Section 3.2). The network dynamics is simulated for 1 s of biological time; spikes are communicated across MPI processes every 1.5 ms. Time is spent on spike delivery (red bars), communication of spike data (yellow bars), neuronal update (green bars), and total sim time (black outline). Error bars (for most numbers of MPI processes hardly visible) indicate the SD over three repetitions. Timings obtained via manual instrumentation of the respective parts of the source code, measured on JURECA CM (Section 3.3).