Noise Benchmark for Qiskit Aer

This is the benchmark script to evaluate multi-shots optimization for GPU simulation on Qiskit Aer (https://github.com/Qiskit/qiskit-aer). Please install GPU enabled version of Qiskit Aer by pip

$ pip uninstall qiskit-aer
$ pip install qiskit-aer-gpu

or build from source code by enabling GPU support by setting -DAER_THRUST_BACKEND=CUDA to cmake option.

How to run benchmark

noise_bench_qft_qubits.py evaluate QFT simulation with Pauli or Kraus noises for some number of qubits. There are 10 parameters required to run the script.

$ python noise_bench_qft_qubits.py method device noise_model shots num_execution error_ratio max_batchable_qubits max_qubits thread_per_gpu

• method : method to be used for the simulation, statevector or density_matrix can be applied
• device : CPU or GPU can be applied for the simulation
• noise_model : Pauli or Kraus noise model can be applied
• shots : number of shots to be simulated for each number of qubits
• num_execution : number of run for each number of qubits, simulation time is average of all iterations
• error_ratio : error ratio of noise sampling
• max_batchable_qubits : batched multi-shots optimization will be applied to less or equal to this option, set 1 to disable multi-shots optimization or set same as max_qubits to enable optimization
• max_qubits : max number of qubits to be evaluated
• thread_per_gpu : This is used for Pauli noise optimization on GPU, please try some numbers that makes the best performance

Following example runs 4000 shots of statevector simulation for 9-20 qubits with 1% of Pauli noise by GPU and batche multi-shots optimization is enabled.

$ python noise_bench_qft_qubits.py statevector GPU Pauli 4000 10 0.01 30 20 4