Gravitational-wave inference at GPU speed: a bilby-like nested sampling kernel within blackjax-ns
GPU 速度下的引力波推断:blackjax-ns 中类 bilby 的嵌套采样内核
We present a graphics processing unit (GPU)-accelerated implementation of the gravitational-wave Bayesian inference pipeline for parameter estimation and model comparison. Specifically, we implement the `acceptance-walk' sampling method, a cornerstone algorithm for gravitational-wave inference within the bilby and dynesty framework. By integrating this trusted kernel with the vectorized blackjax-ns framework, we achieve typical speedups of 20--40<inline-formula><tex-math>$\times$</tex-math></inline-formula>, when comparing total central processing unit (CPU) and GPU core hours, for aligned spin binary black hole analyses, while recovering posteriors and evidences that are statistically identical to the original CPU implementation. While CPU and GPU core hours are not equivalent, we show nevertheless that the GPU-accelerated analyses are up to 3<inline-formula><tex-math>$\times$</tex-math></inline-formula> more cost-effective than their CPU counterparts. This faithful re-implementation of a community-standard algorithm establishes a foundational benchmark for gravitational-wave inference. It quantifies the performance gains attributable solely to the architectural shift to GPUs, creating a vital reference against which future parallel sampling algorithms and machine learning based methods can be rigorously assessed. This allows for a clear distinction between algorithmic innovation and the inherent speedup from hardware. Our work provides a validated community tool for performing GPU-accelerated inference on gravitational-wave data, and demonstrates the scaling potential of nested sampling.
展开 ▾把 bilby/dynesty 中成熟的 acceptance-walk 内核移植到 GPU 向量化 blackjax-ns,典型核时加速 20–40 倍且结果统计一致;相较 CPU 可节省约 3 倍成本,为区分硬件加速与算法创新提供定量基准。
该工作承接引力波贝叶斯推断中已被广泛验证的 acceptance-walk 嵌套采样内核,这一内核此前在 CPU 框架中是社区标准。通过将其移植到 GPU 向量化实现,文章测定了仅由硬件架构迁移带来的加速上限,并建立可复现的基准。这使未来并行采样算法和机器学习方法能够在同等 GPU 硬件条件下被严格评估,避免把硬件增益误判为算法创新。同时,该工具也为更大规模嵌套采样和引力波数据的近实时分析提供了可能性。
刊出 2026-04-29 · 收录 2026-09-14