Efficient implementation of the overlap operator on multi-GPUs

Abstract

Lattice QCD calculations were one of the first applications to show the potential of GPUs in the area of high performance computing. Our interest is to find ways to effectively use GPUs for lattice calculations using the overlap operator. The large memory footprint of these codes requires the use of multiple GPUs in parallel. In this paper we show the methods we used to implement this operator efficiently. We run our codes both on a GPU cluster and a CPU cluster with similar interconnects. We find that to match performance the CPU cluster requires 20-30 times more CPU cores than GPUs.

8 pages with 10 figures; accepted for presentation at the 2011 Symposium on Application Accelerators in High Performance Computing (Knoxville, July 19-20, 2011)

References in corpus (7)

• Solving Lattice QCD systems of equations using mixed precision solvers on GPUs
• Lattice QCD as a video game
• Numerical techniques for lattice QCD in the $\epsilon$--regime
• On the low fermionic eigenmode dominance in QCD on the lattice
• A note on Zolotarev optimal rational approximation for the overlap Dirac operator
• Minimizing storage in implementations of the overlap lattice-Dirac operator
• A note on Neuberger's double pass algorithm

Cited by in corpus (5)

• Scaling Lattice QCD beyond 100 GPUs
• Charmed and $\varphi$ meson decay constants from 2+1-flavor lattice QCD
• QCDGPU: open-source package for Monte Carlo lattice simulations on OpenCL-compatible multi-GPU systems
• Electric polarizability of hadrons with overlap fermions on multi-GPUs
• Locality and Efficient Evaluation of Lattice Composite Fields: Overlap-Based Gauge Operators