Some technological aspects of GPGPU applications in applied program systems

The problem of natural microseismic field analysis is considered. The performance of GPGPU technology is estimated when using single or multiple GPUs and various programming technologies, such as CUDA, JCuda and OpenCL. A comparison is made with the performance of the distributed memory cluster MVS-100K. The influence of several CUDA optimization techniques is estimated. The performance of a GPU cluster is studied with the use of several interprocessor communication technologies (MPI, MPJ, GridGain).