Over the past year and a half, the creation of clusters and supercomputers on the GPU has become a new trend for high-performance systems. This makes perfect sense, because GPUs perfectly support parallel operation, and on some tasks they have much higher performance than budget CPUs.
In this sense, the news looks quite logical, that the native support for Nvidia CUDA (Compute Capability version 1.3 or higher) appeared in the MATLAB version 2010b mathematical modeling software package. It is included in the Parallel Computing Toolbox . Now the parallelization of commands is performed directly from MATLAB and there is no need to rewrite the code in C ++ or Fortran specifically for CUDA.