Jul 26 – 30, 2021
US/Eastern timezone

Gauge field compression in SU(N) theories and spatial correlations on the lattice

Jul 28, 2021, 2:15 PM
15m
Oral presentation Algorithms (including Machine Learning, Quantum Computing, Tensor Networks) Algorithms (including Machine Learning, Quantum Computing, Tensor Networks)

Speaker

Dean Howarth (Boston University)

Description

A long standing problem associated with performing lattice gauge theory calculations on GPU hardware is latency for both global memory transfers and MPI data transfers. Mitigating these latencies with data compression techniques can vastly improve the performance of solvers and help to combat strong scaling. In this talk we discuss a new gauge field compression technique in which the SU(N) fields are decomposed into their fundamental representation, and then further compressed using their spatial correlations and the zfp library. Other lattice data types which exhibit spatial correlations can also be compressed in a similar manner with varying efficiency.

Primary authors

Dean Howarth (Boston University) Kate Clark (NVIDIA)

Presentation materials