Fast scalable and low-power quantum circuit simulation on the cluster of GPUs platforms
Ahmadzadeh, Armin,Sarbazi-Azad, Hamid
DOI: https://doi.org/10.1007/s11082-024-07492-3
IF: 3
2024-09-27
Optical and Quantum Electronics
Abstract:Quantum computing is a rapidly evolving computational means that offers significant speedups for a variety of scientific applications, including machine learning, unsorted database queries, cryptography, and number factorization. Quantum computing holds a significant advantage in its capacity to run quantum algorithms that surpass current classical computer algorithms. However, practical realization of quantum computers remains in its nascent stages, necessitating reliance on classical computation platforms for simulating quantum circuit behavior and developing quantum algorithms. Simulating quantum algorithms presents a formidable challenge due to exponential memory and computation resource demands. In this study, we introduce a novel approach to optimize computational workload distribution during simulation by leveraging Dynamic Load Partitioning (DLP) between host CPUs and GPUs, with the primary aim of reducing resource and computation time requirements. Our method employs a hybrid CPU-GPU platform that divides quantum circuit simulation into two sections, utilizing parallel vector and recursive methods. We conducted our experiments on a cluster of nodes with multiple GPUs to scale qubit simulation and achieve significant speedup over existing simulators. Specifically, our approach demonstrates a remarkable 96X speedup over recursive path-summing on a single GPU and 12.98X speedup over the latest implementation on a multi-node cluster system. Furthermore, our method exhibits energy efficiency, surpassing the state-of-the-art technique by a 55X improvement. These results underscore the effectiveness of the proposed method, especially in leveraging low-cost systems for simulating and processing quantum circuits.
engineering, electrical & electronic,optics,quantum science & technology