Conveners
Software Development and Machines
- Frank Winter (Jefferson Lab)
Software Development and Machines
- Johannes Heinrich Weber (Humboldt University of Berlin)
Software Development and Machines
- Alexei Strelchenko (FNAL)
I give a status update for the Grid Python Toolkit software project.
The rise of exascale supercomputers has fueled competition among GPU vendors, requiring lattice QCD practitioners to write code that supports multiple GPU architectures and APIs. We present SIMULATeQCD, a simple multi-GPU lattice code for large-scale QCD calculations, mainly developed and used by the HotQCD collaboration. Our open source code is built on C++ and MPI, includes CUDA and HIP...
In the rapidly changing hardware landscape of high performance computing (HPC), binding workforce to optimize simulation software for just a single architecture becomes a sustainability issue.
In this work I explored the feasibility of using performance portable parallel code for a staggered fermion kernel. Fusing the Kokkos C++ Performance Portability EcoSystem with MPI allows to scale on...
In this presentation, we will discuss the application of recent and emerging C++ features, with a focus on portable parallel programming in lattice QCD. Specifically, the discussion will center around certain key features introduced in the C++17, C++20, and C++23 standards, as well as an exploration of some experimental features currently under development.
A primary emphasis will be placed...
I will give a status update of using OpenMP target offloading in the Grid library. As part of the US Exascale Computing Project, we have been investigating the possibility of using a portable programming model in Grid to support execution on different architectures. OpenMP, a directives-based programming model, supports both CPU multithreading and different GPU architectures through...
Deep Learning Models in the Machine Learning Community relies heavily on GPU-based tensor calculations. In recent years, Tensor Networks Methods have been explored to estimate the Partition Function of a system deterministically. One of the reasons Tensor Networks have not been yet utilised to the maximum potential in the Lattice Gauge Theories is their time complexity issue of the algorithms....
The Karsten-Wilczek action is a formulation of minimally doubled fermions on the lattice which explicitly breaks hypercubic symmetry and introduces three counterterms with respective bare parameters. We present a tuning of the bare parameters of the Karsten-Wilczek action on 4-stout configurations at the physical point.
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...
The application of normalizing flows for sampling in lattice field theory has garnered considerable attention in recent years. Despite the growing community at the intersection of machine learning and lattice field theory, there is currently a lack of a software package that facilitates efficient software development for new ideas in this field. We present the idea of NeuLat, a fully...
The scientific method is underpinned by reproducibility, however, parallel computing often violates this through lack of associativity when summing floating point numbers. For Lattice QCD calculations this can have several undesirable effects, such as dramatic variations in solver iteration count, as well as the fundamental inability to exactly reproduce a given Monte-Carlo generated...
In the current climate and energy crisis context, it is crucial to study and optimise the energy efficiency of scientific software used at large scale computing facilities. This supports moving toward net-zero computing targets, and reduce the negative impact of growing operational costs on the production of scientific data.
The energy efficiency of a computation is generally quantified as an...