Tribrat M. Investigation the ways of improving the productivity of cluster systems using video adapters

The dissertation proposes a method of increasing productivity and efficient use of cluster systems using video cards to perform a calculation of wide application and implementation of this method in the construction or upgrading of cluster systems. Feasibility of using a video adapter as a CPU coprocessor to perform calculations of general purpose (technology GPGPU) is shown also illustrates the use of hardware-software platform CUDA and video adapters of NVIDIA company. V.M. Glushkov's mathematical apparatus of modified systems of algorithmic algebras is applied to create the formalized scheme of parametric algorithms. The equivalent transformation of developed formalized parametric schemes design of parallel applications made for node of cluster systems (SMP-architecture) and cluster systems (MPP-architecture) with a video adapter. Scalability of resources is confirmed by experiment and is shown through the use of multiple adapters with simultaneous use of hardware-software platform CUDA and technology OpenMP in a single node cluster system.