Tutorial on CUDA parallel programming technologies
February 15, 2017
room 407, LIT JINR
(once a week during 4 weeks)
The following issues on parallel programming applications for carrying out computations using NVIDIA GPUs will be presented in the course of the tutorial:
– tha basics of CUDA;
– OpenMP+CUDA hybrid technology for multi-GPU computations;
– use of matjematical libraries optimized for NVIDIA GPU: cuBLAS, cuFFT, cuSPARSE.
Numerical solution of the initial boundary value problem for heat multidimensional equation of locally one approach will be discussed in respect to CUDA-application development using cuSPARSE library functions for solution of SLAE.