What is NUWA?
NUWA is the Digital Research Instrument (IRN) of the Laboratoire d’Aérologie (LAERO) and the Observatoire Midi-Pyrénées (OMP). It is a high-performance computing and storage cluster, designed to meet the needs of numerical modeling and data processing research. NUWA consists of a heterogeneous computing cluster with 144 compute nodes (2,800 cores), including 7 GPU-equipped nodes (NVIDIA Titan, Tesla V100, Hopper H100), and a storage cluster with a capacity of 2.5 PB. NUWA is primarily funded by research projects (LAERO, OMP, regional, ANR, and European grants). As of 2025, 183 users were registered, mainly from LAERO, LEGOS, GET, IRAP, and external institutions.
Unlike regional, national, and international centers, NUWA supports long simulations (>15 days) and custom reservations on compute nodes. It also serves as a dedicated machine for real-time support of atmospheric and oceanic observation campaigns—a mission incompatible with traditional computing infrastructures due to data flow constraints. Its heterogeneous architecture allows for the deployment of innovative compute nodes, still unavailable on large-scale centers. This ensures active technological monitoring and facilitates porting to more powerful infrastructures.
NUWA is a versatile, flexible, and shared tool, essential for numerical modeling and data processing research at LAERO and OMP. It combines computing power, massive storage, and innovative services, while remaining accessible and adaptable to user needs. (Photo taken on January 7, 2026.)
What does LAERO do for NUWA?
NUWA is a cluster designed and administered by LAERO. The team ensures not only its deployment and maintenance, but also regular updates to guarantee optimal performance and compatibility with the latest technologies.
What is NUWA used for at LAERO?
At LAERO, NUWA is primarily used for:
- Code development and optimization: Development, prototyping, and debugging of numerical codes (such as Meso-NH or CROCO), as well as GPU porting and optimization for modern architectures.
- Code porting: A test platform to facilitate access to regional, national, and international computing centers.
- Real-time support for atmospheric and oceanic observation campaigns: Design and provision of dedicated numerical simulations for campaigns.
- Data processing: Analysis of observation campaigns, satellite data, and forecasts.
- Education: A training tool for high-performance computing (HPC) for students.
- Advanced visualization: Remote visualization on compute nodes (via VirtualGL on GPUs and JupyterLab), especially for environments requiring large memory.
Link to NUWA documentation:
NUWA’s History and Infrastructure
- 2000: NUWA’s predecessor, built around the Meso-NH model, with 12 PCs in a Beowulf network.
- 2010: First NUWA cluster with 104 nodes (496 cores) and 100 TB of storage.
- 2019-2020: Expansion to 154 nodes (2,000 cores) and 900 TB of storage, followed by migration to a 1.5 PB storage cluster.
- 2025: NUWA now includes a heterogeneous computing cluster with 144 compute nodes (2,800 cores), including 7 GPU-equipped nodes (NVIDIA Titan, Tesla V100, Hopper H100), and a 2.5 PB storage cluster.