TARGETS.md

CUDA Quantum target

A target is a specification of the desired platform and simulator / QPU. It can be specified in C++ as a compile time flag in C++ and in Python as a runtime flag. Alteratively, it can also be specified within the application code.

Simulation backends

• state-vector (cuStateVec)
• density-matrix (dm)
• tensor-network (cuTensorNet)

Hardware backends

• CPU only
• Single GPU
• Multi-GPU
• Multi-QPU
• Multi-node

Targets

• default
  • qpp, CPU-based and multithreaded
    • In C++, nvq++ ghz_state.cpp -o a.out
    • In Python, python3 ghz_state.py
• nvidia
  • custatevec with single-GPU
    • In C++, nvq++ --target nvidia ghz_state.cpp -o a.out
    • In Python, python3 --target nvidia ghz_state.py
• nvidia-mqpu
  • custatevec with multi-QPU

    $ python --target nvidia-mqpu distribute_hamiltonian.py
    

• cuquantum_mgpu
  • custatevec with multi-GPU
    • Not covered in this workshop

    $ nvq++ --target cuquantum-mgpu ghz.py -o a.out
    $ mpiexec -np 2 ./a.out
    

• density-matrix-cpu
  • CPU-only multithreaded density matrix emulation
• quantinuum
  - --target quantinuum
• ionq
• more!

To print a list of all the targets in Python:

import cudaq

targets = cudaq.get_targets()

for t in targets:
     print(t)

Note: Some of these targets are not available for this workshop.