Benchmarking Quantum Simulators for Grover’s Algorithm in SCAYLE: A Comparative Study with Classical Systems

Abstract

This paper presents a comparative performance study of Grover’s quantum search algorithm executed through classical emulation across multiple quantum programming frameworks: Qiskit, MyQLM, PennyLane, and Qulacs, on heterogeneous high-performance computing (HPC) architectures provided by SCAYLE. In this work, the search space is defined as a synthetic, uniformly distributed dataset of size 2^n, encoded directly in the computational basis of an n-qubit register. A single marked element, specified as an integer in the interval [0, 2^n - 1], is embedded into the oracle, which inverts its phase. The number of Grover iterations is set to [pi/4,sqrt(2^n)], ensuring optimal amplitude amplification. The same circuit structure is implemented in all frameworks and executed on multiple HPC infrastructures: two CPU-based nodes built on Intel Xeon Platinum 8358 (Ice Lake) and Intel Xeon Platinum 8462Y (Sapphire Rapids), and an NVIDIA Grace CPU-based architecture. Memory configurations range from 256GB to 2TB per node. By varying the number of qubits and execution environment, we evaluate execution time, memory consumption, and scalability limits, providing a systematic comparison of software stacks and processor generations. Notably, the Grace-based node demonstrates a remarkable acceleration in emulation, particularly for Qiskit and PennyLane, highlighting the impact of high memory bandwidth and modern CPU architectures on large-scale quantum algorithm simulation.