Simulation of large neuronal networks in cloud and grid with graphics processing units

Software for simulation of large networks of coupled non-linear oscillators in clusters, grids and clouds using graphical processing units (GPU) was designed, developed, tested and applied for scientific simulations. The software provides easy integration of new oscillators' models support, dynamic load distribution between hosts' central processing units (CPU) and several GPU devices. Different GPU devices provide speed-up of 12–50 compared to single core Intel Xeon, 2.4 GHz depending on GPU and job types. The software was efficiently applied for simulations of 3D networks with 10 7 –10 8 oscillators described by Kuramoto-Sakaguchi model and new types of “chimera states” were discovered in such simulations.