Assessment of ESM Readiness Level for Exascale HPC Authors

Oliver Elbert; Jessie  Carmin; Frank Giraldo; Mark Govett; Lucas Harris; Thomas Hauser; David McCarren; Joseph  Mouallem; Mark Olsen; Todd Ringler; Sarat Sreepathi; Mark A. Taylor

Assessment of ESM Readiness Level for Exascale HPC Authors

Oliver Elbert
, Jessie Carmin
, Frank Giraldo
, Mark Govett
, Lucas Harris
, Thomas Hauser
, David McCarren
, Joseph Mouallem
, Mark Olsen
, Todd Ringler
, Sarat Sreepathi
, Mark A. Taylor

Sandia National Laboratories, New Mexico
Sandia National Laboratories

Research output: Book or Report › Technical report

Abstract

● U.S. current and upcoming Exascale machines are GPU based.
● Exascale readiness requires porting ESMs to GPUs, involving large efforts: either new code or major refactoring of existing code.
● No dominant programming model to date. Approaches include rewriting in Domain Specific Languages, C++ with template based performance portability libraries and refactoring existing Fortran code to support GPU directives.
● Several atmosphere models are GPU-ready, with some progress on other components. No full ESM is GPU-ready as of 2024.
● Preliminary results from one of the first GCRMs to run on an exascale system show that modern GPU nodes can be ~6x faster than modern CPU nodes, or ~3.5x faster per Watt.
● We did not attempt to estimate GPU performance vs CPU performance on a per cost basis.

Original language	American English
State	Published - May 1 2025

Access to Document

https://www.osti.gov/servlets/purl/2566651

Cite this

@book{d527ce879b58440780ea32c54e171ef2,

title = "Assessment of ESM Readiness Level for Exascale HPC Authors",

abstract = "● U.S. current and upcoming Exascale machines are GPU based. ● Exascale readiness requires porting ESMs to GPUs, involving large efforts: either new code or major refactoring of existing code. ● No dominant programming model to date. Approaches include rewriting in Domain Specific Languages, C++ with template based performance portability libraries and refactoring existing Fortran code to support GPU directives. ● Several atmosphere models are GPU-ready, with some progress on other components. No full ESM is GPU-ready as of 2024. ● Preliminary results from one of the first GCRMs to run on an exascale system show that modern GPU nodes can be \textasciitilde{}6x faster than modern CPU nodes, or \textasciitilde{}3.5x faster per Watt. ● We did not attempt to estimate GPU performance vs CPU performance on a per cost basis. ",

author = "Oliver Elbert and Jessie Carmin and Frank Giraldo and Mark Govett and Lucas Harris and Thomas Hauser and David McCarren and Joseph Mouallem and Mark Olsen and Todd Ringler and Sarat Sreepathi and Taylor, \{Mark A.\}",

year = "2025",

month = may,

day = "1",

language = "American English",

}

TY - BOOK

T1 - Assessment of ESM Readiness Level for Exascale HPC Authors

AU - Elbert, Oliver

AU - Carmin, Jessie

AU - Giraldo, Frank

AU - Govett, Mark

AU - Harris, Lucas

AU - Hauser, Thomas

AU - McCarren, David

AU - Mouallem, Joseph

AU - Olsen, Mark

AU - Ringler, Todd

AU - Sreepathi, Sarat

AU - Taylor, Mark A.

PY - 2025/5/1

Y1 - 2025/5/1

N2 - ● U.S. current and upcoming Exascale machines are GPU based. ● Exascale readiness requires porting ESMs to GPUs, involving large efforts: either new code or major refactoring of existing code. ● No dominant programming model to date. Approaches include rewriting in Domain Specific Languages, C++ with template based performance portability libraries and refactoring existing Fortran code to support GPU directives. ● Several atmosphere models are GPU-ready, with some progress on other components. No full ESM is GPU-ready as of 2024. ● Preliminary results from one of the first GCRMs to run on an exascale system show that modern GPU nodes can be ~6x faster than modern CPU nodes, or ~3.5x faster per Watt. ● We did not attempt to estimate GPU performance vs CPU performance on a per cost basis.

AB - ● U.S. current and upcoming Exascale machines are GPU based. ● Exascale readiness requires porting ESMs to GPUs, involving large efforts: either new code or major refactoring of existing code. ● No dominant programming model to date. Approaches include rewriting in Domain Specific Languages, C++ with template based performance portability libraries and refactoring existing Fortran code to support GPU directives. ● Several atmosphere models are GPU-ready, with some progress on other components. No full ESM is GPU-ready as of 2024. ● Preliminary results from one of the first GCRMs to run on an exascale system show that modern GPU nodes can be ~6x faster than modern CPU nodes, or ~3.5x faster per Watt. ● We did not attempt to estimate GPU performance vs CPU performance on a per cost basis.

M3 - Technical report

BT - Assessment of ESM Readiness Level for Exascale HPC Authors

ER -

Assessment of ESM Readiness Level for Exascale HPC Authors

Abstract

Access to Document

Fingerprint

Cite this