We present a new parallel algorithm for supercomputer simulation of gas-dust circumstellar gravitating disks. The algorithm uses the domain decomposition technique and combines numerical methods of smooth particle hydrodynamics (SPH), particle-in-cell (PIC) and grid-based gravitational solver with the convolution method and parallel multidimensional Fast Fourier Transform.

The algorithm is designed to address the specificity of simulating non-stationary rotating disks, where the matter, represented by SPH and PIC particles, moves across a computational domain, and therefore requires efficient load-balancing methods. The parallel algorithm is suitable for running on supercomputers with different types of architectures: standard (CPU-based), and hybrid (CPU with NVIDIA GPU and Intel Xeon Phi).