Spin Hall and spin transfer torque oscillators hold a great promise for the future rf signal generation and neuromorphic computing technologies. Both applications rely on the ability of these highly non-linear non-identical devices to mutually synchronize once they put together in planar and 3D arrays. Our group was always at the frontier of this research topic by demonstrating experimentally the highest number of simultaneously synchronized oscillators of various, technology-relevant designs. Nevertheless, despite the ongoing efforts of world-leading spintronic groups, the synchronization of the larger number of oscillators have not been reported so far neither experimentally nor in micromagnetic simulations. This is typically attributed to the relatively short range (nearest-neighbor) interaction of such oscillators, their intrinsic frustration, and phase noise. From the simulations point of view, such problems are too computationally intensive and cannot be simply addressed on a reasonable timescale using conventional CPU-based solvers. However, the emergence of the GPU-accelerated micromagnetic codes made it possible to investigate such a large-scale problems using even consumer-grade hardware. So, here we propose to employ micromagnetic simulations to study collective dynamics of large-scale planar arrays and stacks of spin Hall nano-oscillators using the GPU-enabled nodes of the Triolith supercomputer. In particular, we would like to investigate existence criteria of the globally synchronized state (when there is constant phase relation between any given pair of the oscillators withing the array). Then we would like to investigate how this state could be manipulated with external stimuli, e.g. rf magnetic field or electrical current. Finally, we would like to explore how the perturbations of the oscillators frequency, phase and amplitude propagate in such large-scale systems.