We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

physics.comp-ph

Change to browse by:

References & Citations

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo

Physics > Computational Physics

Title: VegasFlow: accelerating Monte Carlo simulation across multiple hardware platforms

Abstract: We present VegasFlow, a new software for fast evaluation of high dimensional integrals based on Monte Carlo integration techniques designed for platforms with hardware accelerators. The growing complexity of calculations and simulations in many areas of science have been accompanied by advances in the computational tools which have helped their developments. VegasFlow enables developers to delegate all complicated aspects of hardware or platform implementation to the library so they can focus on the problem at hand. This software is inspired on the Vegas algorithm, ubiquitous in the particle physics community as the driver of cross section integration, and based on Google's powerful TensorFlow library. We benchmark the performance of this library on many different consumer and professional grade GPUs and CPUs.
Comments: 6 pages, 5 figures, final version published in CPC
Subjects: Computational Physics (physics.comp-ph); High Energy Physics - Experiment (hep-ex); High Energy Physics - Phenomenology (hep-ph); Machine Learning (stat.ML)
DOI: 10.1016/j.cpc.2020.107376
Report number: TIF-UNIMI-2020-8
Cite as: arXiv:2002.12921 [physics.comp-ph]
  (or arXiv:2002.12921v2 [physics.comp-ph] for this version)

Submission history

From: Stefano Carrazza [view email]
[v1] Fri, 28 Feb 2020 18:48:16 GMT (43kb,D)
[v2] Wed, 20 May 2020 10:52:37 GMT (43kb,D)

Link back to: arXiv, form interface, contact.