Many Cores, Many Models: GPU Programming Model vs. Vendor Compatibility Overview

Herten, Andreas

doi:10.1145/3624062.3624178

Cited by 3 publications

(1 citation statement)

References 16 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…While GPU porting may be more feasible using the eDSL approach for the C/C++ language [14], the codes (especially legacy codes) written in Fortran are less supported when using hardware from other vendors than NVIDIA, e.g. GPUs from AMD.…”

Section: Discussionmentioning

confidence: 99%

Challenges of applying an embedded domain specific language for performance portability to Earth system models

Lu,

Caviedes-Voullième,

Stein

et al. 2024

Preprint

View full text Add to dashboard Cite

Weather and climate modeling, reliant on substantial computational resources, faces challenges of escalating resource demands and energy consumption as problem sizes and model complexity increase. Leveraging Graphics Processing Units (GPUs) for accelerated simulations demands performance portability across diverse High Performance Computing (HPC) architectures. The concept of embedded Domain Specific Language (eDSL) emerges as a lightweight solution, aiming to streamline GPU utilization without extensive code rewrites and to enhance code portability across HPC architectures. This study explores the practical implementation of an eDSL within the Icosahedral Nonhydrostatic (ICON) weather and climate model, primarily written in Fortran. Our evaluation contrasts eDSL utilization within ICON against other Earth system models, with particular emphasis on three-dimensional mass transport, or advection. Through assessing the eDSL's efficectiveness in various models, especially its application in the ParFlow and EULAG models, insights into its potential for ICON emerge. While the eDSL presents a promising avenue for performance portability, challenges in adapting Fortran-based codes, GPU support, and resource allocation underscore the need for necessity for thorough planning and resource allocation in model development endeavors. Overall, the eDSL offers a viable pathway for harnessing GPU acceleration while mitigating the complexities of code portability across diverse HPC architectures, essential for advancing weather and climate modeling capabilities.

show abstract

Section: Discussionmentioning

confidence: 99%

Challenges of applying an embedded domain specific language for performance portability to Earth system models

Lu,

Caviedes-Voullième,

Stein

et al. 2024

Preprint

View full text Add to dashboard Cite

show abstract

Very-Large-Scale GPU-Accelerated Nuclear Gradient of Time-Dependent Density Functional Theory with Tamm–Dancoff Approximation and Range-Separated Hybrid Functionals

Kim,

Jeong,

Weisburn

et al. 2024

J. Chem. Theory Comput.

View full text Add to dashboard Cite

Modern graphics processing units (GPUs) provide an unprecedented level of computing power. In this study, we present a high-performance, multi-GPU implementation of the analytical nuclear gradient for Kohn−Sham time-dependent density functional theory (TDDFT), employing the Tamm− Dancoff approximation (TDA) and Gaussian-type atomic orbitals as basis functions. We discuss GPUefficient algorithms for the derivatives of electron repulsion integrals and exchange−correlation functionals within the range-separated scheme. As an illustrative example, we calculate the TDA-TDDFT gradient of the S 1 state of a full-scale green fluorescent protein with explicit water solvent molecules, totaling 4353 atoms, at the ωB97X/def2-SVP level of theory. Our algorithm demonstrates favorable parallel efficiencies on a high-speed distributed system equipped with 256 Nvidia A100 GPUs, achieving >70% with up to 64 GPUs and 31% with 256 GPUs, effectively leveraging the capabilities of modern high-performance computing systems.

show abstract

Unified Schemes for Directive-Based GPU Offloading

Miki,

Hanawa

2024

IEEE Access

View full text Add to dashboard Cite

Many Cores, Many Models: GPU Programming Model vs. Vendor Compatibility Overview

Cited by 3 publications

References 16 publications

Challenges of applying an embedded domain specific language for performance portability to Earth system models

Challenges of applying an embedded domain specific language for performance portability to Earth system models

Very-Large-Scale GPU-Accelerated Nuclear Gradient of Time-Dependent Density Functional Theory with Tamm–Dancoff Approximation and Range-Separated Hybrid Functionals

Unified Schemes for Directive-Based GPU Offloading

Contact Info

Product

Resources

About