CUDA Toolkit

Versions Installed

Kay: 10 / 10.1.243 / 11.2 / 11.3 / 11.4


The NVIDIA CUDA Toolkit provides command-line and graphical tools for building, debugging and optimizing the performance of applications accelerated by NVIDIA GPUs, runtime and math libraries, and documentation including programming guides, user manuals, and API references.


The CUDA toolkit is available under the CUDA Toolkit End User License Agreement (see CUDA Toolkit documentation).



Job Submission Example

Like other jobs on ICHEC systems, CUDA Toolkit jobs must be submitted using a Slurm submission script. The following is an example Slurm submission script for allocating 1 node of Kay (40 cores) for 30 minutes, then running a CUDA-compiled tutorial program matmul, which multiplies large matrices using gpu acceleration.  Since CUDA takes advantage only of gpus, all CUDA jobs must be submitted to the GpuQ.

Load the desired version of the toolkit module:

module load cuda/$version

Following the ICHEC tutorial, we compile the CUDA source file using nvcc as follows:

nvcc -o matmul -arch=sm_70 -O3

Create the submission script (and name it for the SLURM Workload Manager and modify it with your parameters

#SBATCH -t 00:30:00
#SBATCH --job-name=myjobname

# Charge job to myproject
#SBATCH -A myproject

# Write stdout+stderr to file
#SBATCH -o output.txt

# Mail me on job start & end
#SBATCH --mail-type=BEGIN,END


# Load the module for the specific version
module load cuda/$version

./matmul 10000

Submit the job with the following command


Additional Notes

Further information can be obtained from the CUDA Toolkit website

Supported By

File Browser Reference
Department FHERIS
University of Galway
HEA Logo