FIGURE 4.
Computational time in the cerebellar network model. (A) Weak scaling property. We varied the number of compute nodes: 1,024, 4,096, 10,000, 40,000, and 82,944. The horizontal axis is the number of compute nodes, and the vertical axis is the computational time (s) spent for 1 s simulation. The total computational time (blue square) = calculation time of membrane potentials (neuron computational time, red circle), calculation time of synaptic inputs (synapse computational time, yellow circle), and calculation time of communication between nodes (communication time, green circle). With 1,024 and 4,096 nodes, we carried out five simulations. For each node, we obtained the total computational time of 456±52.4, 459±45.2, 429, 425, and 578 s, respectively (blue squares). The neuron computational times were 150±3.03, 154±14.6, 149, 148, and 150 s, respectively. The synapse computational times were 223±3.01, 224±2.23, 237, 229, and 225 s, respectively. The communication times were 83±54, 80±40, 43, 48, and 202 s, respectively. The data for 1,024 and 4,096 nodes are displayed as mean ± standard deviation. The horizontal axis is shown in log-scale. (B) A pie chart of the components of computational time with 1,024 nodes. (C) A pie chart of the components of computational time with 1,024 nodes for the cerebellar network model with spontaneous discharge.