CUDA is a parallel computing platform and programming model invented by NVIDIA for GPUs (graphic processing unit).

Category:  Compiler

Availability: Open access for all academic users

The CUDA toolkit is available for use on nodes that have the GPU devices.  The necessary NVidia device drivers are installed on these nodes.  The 12-core nodes with GPUs are in both the general-compute partition and the gpu partition.  Specify the gpu partition and 'gres' flag when submitting jobs that will use the GPUs.   One 16-core node with 2 GPUs is available in the debug partition.   See this KB article for SLURM flags to request these nodes.

Usage Notes

Show the software versions:  module avail cuda

Loading the module will set the path and any necessary variables:   module load cuda/version

Extensive documentation on CUDA programming using C/FORTRAN and other languages is available from NVIDIA's CUDA Developer Zone.