EPIC has successfully coordinated the porting of the Unified Forecast System (UFS) Weather Model (WM) to NOAA’s new high-performance computing system, Ursa. As NOAA’s Tier-1 Research and Development High Performance Computing System (RDHPCS), Ursa is located at the NOAA Environmental Security Computing Center (NESCC) in Fairmont, West Virginia, and will support advanced weather modeling to enhance the prediction of high-impact weather events. Ursa’s Graphics Processing Units (GPUs) will help to boost NOAA’s growing use of Artificial Intelligence (AI) and Machine Learning (ML) in weather modeling, data assimilation, and predictive analysis.
Ursa is built on Advanced Micro Devices, Inc (AMD) Genoa processors, featuring:
- 576 compute nodes, each with 2 sockets and 96 cores per socket (192 cores per node), totaling 110,592 CPU cores and delivering 4.25 PFlops peak performance.
The porting effort was a collaborative initiative involving the EPIC Software Integration (SI), the UFS WM and Applications Code Management (CM), and the Spack Stack Library Teams, with support from the RDHPC team. This interdisciplinary partnership ensured a smooth transition to Ursa, with full alignment to the UFS WM regression testing requirements, enabling reliable and scalable performance for operational forecasting test cases. The porting process was completed by establishing full compatibility between the UFS WM codebase and Ursa’s architecture, utilizing the Spack Stack version 1.9.2 release. This involved adapting the model to Ursa’s compiler suite, MPI libraries, and file system structure, ensuring a stable and efficient execution environment. The performance comparison (Figure1) illustrates how the UFS WM regression tests (RTs), represented by points along the X-axis, performed on the Ursa system, with results benchmarked against those from current platform, Hera.
In most test cases, Ursa achieved a significant reduction in wall-clock time. For instance, the cpld_control_gfsv17 test case when using the Intel compiler showed a reduction of approximately 30 – 40% compared to the same run on Hera. While Ursa consistently improves the wall-clock time for RT cases overall, sporadic spikes in wall-time have been observed in some individual testcases. As it is a newly deployed machine, optimization efforts are currently underway and are expected to mitigate these wall-time fluctuations over time.
This successful porting of the UFS WM to Ursa marks a significant advancement in NOAA’s modeling capabilities, by enabling faster, more scalable, and more accurate simulations to support critical weather prediction efforts. The collaboration between EPIC, UFS, and Spack Stack teams demonstrates the power of coordinated software integration and high-performance computing innovation.
For questions or further information about the porting process or performance benchmarking, please contact the EPIC User Support and Community Engagement teams at support.epic@noaa.gov.
