CUDA-Q

CUDA-Q is a software development kit for quantum computing from NVIDIA. Additional training materials can also be found at CUDA-Q Academic.

CUDA-Q is available in Midway3 as a Singularity file (which has both the C++ version and the Python version) and also as a separate Python environment via the software module cudaq. As with other software packages, when using CUDA-Q for computation, the compute nodes should be used, which could be accessed via interactive sessions or batch jobs.

The path to the Singularity file and the Python environment can be obtained from the module details via the following command.

module show cudaq

The Python environment is available at CUDAQ_ENV and the Singularity file at CUDAQ_SIF. The Python environment could be activated using source $CUDAQ_ENV after loading the module module load cudaq/.

To use the Singularity file for CUDA-Q, you can load the Singularity module, using module load singularity/. And then to use CUDA-Q, you could either use the Singularity shell or execute the program directly. Instructions on using Singularity in Midway3, can be found at this page.

For the first approach, you can launch the Singularity shell usingsingularity shell $CUDAQ_SIF, and then run the Python program e.g., using python example_program.py or the C++ program by compiling the code first, e.g., using nvq++ example_program.cpp -o example.out, followed by the execution e.g., using ./example.out.

For the second approach, you could execute Python or C++ program using the exec command, e.g., using singularity exec $CUDAQ_SIF python example_program.py > example_output.txt, for the case of Python.

When using the NVIDIA GPU resources, you can use the --nv option with the singularity file e.g., when launching the Singularity shell, use singularity shell --nv $CUDAQ_SIF. An example batch script for running CUDA-Q jobs on GPU nodes is given below.

#!/bin/bash
#SBATCH --nodes=1
#SBATCH --ntasks-per-node=1
#SBATCH --gres=gpu:1
#SBATCH --account=pi-cnetid
#SBATCH --time=01:00:00
#SBATCH --partition=gpu
#SBATCH --job-name=cudaq_job
#SBATCH --output=output-%J.txt
#SBATCH --error=error-%J.txt

module load singularity/
module load cudaq/

# Run the Python program using CUDA-Q in the Singularity file
singularity exec --nv $CUDAQ_SIF python my_quantum_program.py