Revisiting Quantum Supremacy: Simulating Sycamore-Class Circuits Using Hybrid CPU/GPU HPC Workloads

TL;DR

This paper demonstrates a hybrid CPU/GPU classical simulation of quantum circuits used in quantum supremacy experiments, achieving significant speedups and challenging the notion of fixed quantum advantage.

Contribution

It introduces a hybrid HPC framework combining GPU and CPU resources to simulate large quantum circuits more efficiently than previous CPU-only methods.

Findings

Achieved a high fidelity simulation with XEB score of 0.549.

Completed a 20-cycle circuit simulation in 1 hour 15 minutes, a 6.95×10^7 speedup.

Estimated that with more CPUs, classical simulation could match quantum experiment times.

Abstract

We present a framework for effectively simulating the execution of quantum circuits originally designed to demonstrate quantum supremacy using accessible high-performance computing (HPC) infrastructure. Building on prior CPU-only approaches, our pipeline combines a single NVIDIA A100 GPU for quantum state construction, followed by N parallel CPU jobs that perform distributed measurement sampling. We validate the fidelity by simulating the 53-qubit, 14-cycle Sycamore circuit and achieving a linear cross-entropy benchmarking (XEB) score of 0.549, exceeding the published XEB score of 0.002 from Google's reference data. We then evaluate execution time performance with the more complex 53-qubit, 20-cycle circuit, completing the full 2.5 million-shot workload over 100 CPU jobs in 01:15:36, representing a 6.95 x 10^7 speedup compared to Google's original classical estimate. Further, we show that if 1,000 CPU jobs were employed, the estimated duration would be approximately 00:17:35, only 12 minutes slower than the time taken by the original QPU-based experiment. These results illustrate that 'quantum supremacy' is not fixed and continues to be a moving target. In addition, hybrid classical-quantum strategies may provide broader near-term quantum utility than once thought.