Please read these instructions before posting any event on Fermilab Indico

Indico will be down for maintenance on Thursday, May 16th from 6:00PM - 8:00PM CST.

31 July 2023 to 4 August 2023
America/Chicago timezone

Optimizing Staggered All-to-All for GPUs

1 Aug 2023, 17:20
20m
Curia II (WH2SW)

Curia II

WH2SW

Speaker

Michael Lynch (University of Illinois, Urbana-Champaign)

Description

We present our implementation of the all-to-all meson field and low mode averaging (LMA) calculations, built on the Grid and Hadrons libraries. We discuss code optimizations made for staggered fermions and GPU offloading, as well as benchmark comparisons on leadership-class resources. We conclude with the statistical gains achieved using LMA for vector-current two-point functions relevant for computing the HVP contribution to muon $g-2$. The calculation is performed on 2+1+1 HISQ ensembles at physical pion mass and lattice spacings as small as 0.06 fm. We discuss the effect of this low-mode--improved data set on long-distance noise and the results of combining our new data with our previous high-statistics data.

Topical area Software Development and Machines

Primary author

Michael Lynch (University of Illinois, Urbana-Champaign)

Presentation materials