Reproduce “Scaling Sample-Based Quantum Diagonalization on GPU-Accelerated Systems using OpenMP Offload” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Robert Walkup, Juha Jäykkä, Igor Pasichnyk, Zachary Streeter, Kasia Świrydowicz, Mikko Tukiainen, Yasuko Eckert, Luke Bertels and 9 more
Researchers from IBM, AMD, and ORNL optimized the classical diagonalization step within the Sample-based Quantum Diagonalization (SQD) method for GPU-accelerated systems using OpenMP Offload, achieving up to a 95x speedup per node compared to CPU implementations and reducing computation time for complex molecular systems from hours to minutes.
Key insights
- Substantial GPU acceleration is achievable for complex scientific codes, even those with low floating-point intensity and deep control flow, by adapting them for parallel architectures.
- Introducing a persistent configuration cache was crucial for GPU performance by eliminating redundant bit-string calculations, demonstrating the benefit of trading memory for computation on modern GPUs.
- OpenMP target offload offers a portable and efficient programming model, enabling a single codebase to target diverse GPU architectures from different vendors (AMD, NVIDIA) effectively.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.