CUDA is a programming model for Nvidia graphical processing units (GPU). HPC cluster has six Tesla M2070 and eight Tesla K40 GPU. CUDA C++/Fortran programming libraries are available upon logging into the RTU HPC cluster server (instruction is available here).
Linux-related graphical tools emacs and ddd can be used for programming and debugging, for example:
module load cuda/cuda-7.5
nvcc -g -G test.cu –o test.out
ddd –debugger cuda-gdb test.out
To be able to use GUI tools, X11 forwarding must be performed to provide a remote graphical interface (instruction is available here).
Nvidia-developed and Eiclipse-based development environment can be used by running the command nsight.
Login node ui.hpc.rtu.lv is only intended for developing programs. If executing of CUDA program task requires too many CPU resources, the task should be executed on computing nodes. Description of some simple operations with cluster is available here.