Fujitsu HPC System at Japan’s Meteorological Research Institute

Fujitsu Limited has completed the construction of an HPC system for the Meteorological Research Institute of the Japan Meteorological Agency.

The institute started operating the new system on March 2.

The new system consists of approximately 900 nodes including the latest x86 servers Fujitsu Server PRIMERGY CX2550 M5. Also, by constructing a cluster by connecting servers with a high-speed interconnect Intel Omni-Path Architecture, the system achieves the theoretical computational performance of 2.81 petaflops.¹

Expectations for more accurate weather forecasts are increasing, as natural disasters are expanding and intensifying with abnormal weather, large-scale earthquakes and tsunamis.

Since its establishment in 1946, the institute has been conducting a range of research related to weather and climate forecasting. It is working to develop technologies for predicting torrential downpours and storms associated with typhoons, while addressing issues such as adaptation to warming. R&D are also being carried out on models for simulating conditions ranging from local weather to climate forecasts. Moreover, efforts are being made to upgrade meteorological services, so as to achieve the world’s highest level of weather forecast accuracy by around 2030.

The Institute contributes to socio-economic activities by providing accurate weather information and data, as it supports the safety and security of the population through warnings and information issued by the Japan Meteorological Agency.

The system connects 880 nodes of the latest x86 servers Fujitsu Server PRIMERGY CX2550 M5, equipped with Intel’s newest CPUs 2nd gen Intel Xeon scalable processors, in parallel, using a high-speed interconnect Intel Omni-Path Architecture. Accordingly, it achieves high scalability for simulation applications such as weather forecasting, earthquake and tsunami analysis conducted by the institute, and can perform calculations on a range of applications, from small to large scales.

The system is CPU-based and does not use accelerators such as GPUs, enabling a range of applications to run with low power consumption. The CPU uses a water-cooling method with high cooling efficiency, which can reduce heat gen from the server with less power consumption than air-cooled systems that use cooling fans.

With the new HPC system, the institute plans to further advance R&D in areas such as the prediction and analysis of earthquakes, tsunamis, and volcanoes. Moreover, the system will be used for the development of high-resolution, detailed numerical prediction models as well as for research on predicting warming.

At the same time, with its technology and experience accumulated through the development and offering of HPCs, Fujitsu will support the institute in their research to achieve the world’s highest level of weather forecast accuracy, thereby contributing to realize a safe and secure society.

Images of the technology predicting storms and heavy rain caused by local downpours and typhoons

Main configuration of the new system

“As an institution of the Japan Meteorological Agency, the Meteorological Research Institute conducts research related to the observation, analysis, and prediction of various conditions ranging from the meteorological phenomena, terrestrial phenomena, and hydrological phenomena, while promoting the development of fundamental technology applicable to future tasks of the Agency,” commented Kazutoshi Onogi, senior director for research coordination, Meteorological Research Institute. “A precise numerical prediction model with high resolution is necessary to accurately predict storm and heavy rain associated with typhoons as well as local downpours including linear precipitation zones that cause major disasters, and to provide appropriate disaster prevention information. The new supercomputer system can meet our expectations, and we will use it to further promote our research and development, for example in predicting global warming in addition to advancing analysis and prediction technologies in the fields of earthquakes, tsunamis, and volcanoes.“

¹Petaflops: Short for peta floating point operations per second. Peta is an SI prefix indicating one quadrillion, or 10 to the power of 15, indicating performance of one quadrillion floating-point operations per second.