Figure 4
From: Brian2GeNN: accelerating spiking neural network simulations with graphics hardware
Minimal biological runtime after which the total simulation time, including preparations such as code generation and compilation (cf. Fig. 3), is smaller when using a GPU compared to 24 threads on a CPU, for networks of different sizes. This data was calculated from benchmark results as displayed in Fig. 2. The CPU comparison is the host of the Titan Xp GPU (see Table 1). Results for the Mbody benchmark (left) and the COBAHH benchmark (right). The calculations are based on single precision performance for the Quadro GPU (blue) and Titan Xp GPU (green), and on double precision performance for the Tesla K40c (orange) and the Tesla V100 GPU (purple).
