A hybrid GPU-CPU effective realization of a parallel numerical algorithm has been developed

 

   In the course of the cooperation between IKIR FEB RAS and Vitus Bering Kamchatka State University, a hybrid GPU-CPU effective realization of the parallel numerical algorithm of nonlocal finite-difference scheme for Cauchy problem solution for a nonlinear differential Riccati equation of fractional variable order has been developed. The paper shows a software code of the algorithm in C and CUDA C languages developed during the application of software-hardware OpenMP and CUDA architectures.

   As the result, a paper was published, which presented the investigation of calculation efficiency of the suggested parallel algorithm based on the data from the series of computational experiments using a computational server NVIDIA DGX STATION. It was shown on test examples that the hybrid version of the numerical algorithm can give significant increase of the performance by 3-5 times compared to the algorithm subsequent version.

   The work was carried out within the RF Grant MD-758.2022.1.1 on the subject "Development of mathematical models of fractional dynamics to investigate oscillation processes and processes with saturation», registration number АААА-А21-121011290003-0.

 

Tverdyi D.A., Parovik R.I. Hybrid GPU-CPU efficient implementation of a parallel numerical algorithm for solving the Cauchy problem for a nonlinear differential Riccati equation of fractional variable order \ Mathematics, 2023, vol. 11, issue. 15, p. 1--21, doi: 10.3390/math11153358.