Benchmarking quantum computers for industrial applications

Source: press release Terra Quantum

Terra Quantum, a European quantum technology company, presents a comprehensive report benchmarking publicly available simulated and native quantum computing platforms. The company says this can be used to find the best available option for industrial applications in each case.

Such benchmarking, according to a Boston Consulting Group study released earlier this year (“The Race to Quantum Advantage Depends on Benchmarking”)” an important topic in the industry. The study, “Benchmarking simulated and physical quantum processing units using quantum and hybrid algorithms,” compares the speed, cost and quality of results of publicly available native quantum processors: from Ionq, Rigetti, Oxford Quantum Circuits and IBM, and simulated quantum processors from Qmware and Amazon Braket. The first benchmark was completed on November 25, 2022.

The goal of the study was to investigate how the prediction accuracy and training time of quantum neural networks can be improved by using a hybrid quantum-classical approach. The benchmark found that the combination of simulated quantum processors and high-performance classical computers provided the most powerful, cost-effective, and robust approach.

In the selection

The first and includes a variety of publicly available native (QPUs: “Ionq Harmony”, “Rigetti Aspen M-2”, Oxford Quantum Circuits’ “OQC Lucy” and “IBM Falcon r5.11”) and simulated quantum computers (“AWS m5.24xlarge”, “AWS SV1” and “Qmware HQC4020”. Because the study focused on real-world applications, it deliberately excluded systems that are exclusive or still at an experimental stage, for example, IBM’s “Statevector” simulator and the “Atos Quantum Learning Machine”; QPUs such as IBM’s “Eagle” processor, Honeywell’s “Model H1” system, and Google’s “Sycamore” processor; and quantum simulationbackends such as IBM’s “Qiskit” machine learning suite and the “Qulacs” package. Terra Quantum plans to continue benchmarking and expand the study accordingly once the company believes the technology is mature enough.

Markus Pflitsch, CEO and founder of Terra Quantum, provides the following explanation: “The benchmark provides reliable insights for anyone evaluating the implementation of quantum computing to efficiently take advantage of quantum technology today and prepare for the quantum future.”

With these constraints, the benchmarking measured the combination of speed, cost of ownership and quality of performance. The results show that a combination of simulated quantum processors and classical high-performance computing high-performance computing (hybrid approach) is currently the most time- and cost-efficient solution for training quantum algorithms.

The elimination of problems

For example, training neural networks on a simulated quantum processor avoids the costly repetitions required by native quantum processors due to frequent computational errors. Today’s error-prone native quantum processors reduce the functional capability of large physical qubits to only a few usable qubits.

In fact, overcoming these errors is currently one of the biggest challenges for native quantum processor vendors. Simulated quantum processors currently offer up to 40 error-free, so-called algorithmic qubits – QMware up to 40 and AWS SV1 up to 34 simulated qubits. Georg Gesek, CTO and co-founder of Qmware, explains, “For algorithms requiring less than 30 qubits, they delivered even faster results than their native counterparts.”

He adds, “The QPU simulated by Qmware is to date the fastest publicly available option for algorithms requiring fewer than 27 qubits.” In the benchmark, the vendor’s software and hardware platform processed data 78 percent faster than the second-best, he said.