상세 보기
AURORA-Q: Asynchronous Unified Resource Optimizer for Quantum Simulation on HPC System
- Kim, Changjong;
- Sim, Alex;
- Wu, Kesheng;
- Tang, Houjun;
- Son, Yongseok;
- 외 2명
WEB OF SCIENCE
0SCOPUS
0초록
This paper presents AURORA-Q, a tiered-memory quantum circuit simulation framework designed to overcome the memory bottleneck and scalability limits of large-scale quantum circuit simulation on Perlmutter. Its primary strategy is to extend beyond GPU-centric designs bounded by VRAM and introduce a unified tiered-memory architecture that integrates GPU VRAM, DRAM, and storage into a simulation framework. With this design, AURORA-Q partitions the full state vector into subsets, manages their residency across memory tiers, overlaps computation with data movement through an async execution pipeline, and employs a dynamic two-level cache with adaptive resource control to exploit access locality under constrained resources. Our evaluation demonstrates that AURORA-Q executes up to 43 qubit simulations on 512 GPUs, a scale unattainable by existing state-of-the-art simulators due to limited memory resources. AURORA-Q achieves up to 2.91 × speedup compared to baselines, while maintaining up to a 79.3% cache hit rate and reducing pipeline stalls by 73% to hide I/O latency.
키워드
- 제목
- AURORA-Q: Asynchronous Unified Resource Optimizer for Quantum Simulation on HPC System
- 저자
- Kim, Changjong; Sim, Alex; Wu, Kesheng; Tang, Houjun; Son, Yongseok; Park, Jisung; Kim, Sunggon
- 발행일
- 2026
- 유형
- Proceedings Paper
- 페이지
- 993 ~ 1003
- 언어
- ENG
- 출판사
- Institute of Electrical and Electronics Engineers Inc.
- 분량
- 11 페이지
- ISSN
- E 2575-8411
P 1063-6927