PhysicsNeMo: Datacenter CFD Simulations for Hot-Aisle Thermal Analysis

NVIDIA released this dataset in early 2026 containing normalized OpenFOAM Computational Fluid Dynamics (CFD) simulations for datacenter thermal management. It captures variations in IT rack counts and geometric parameters including aisle length, height, and width under maximum flow rate conditions.

Use Cases

• Training surrogate models to predict thermal fields from aisle length and width parameters
• Benchmarking physics-informed machine learning models against OpenFOAM-solved flow data
• Sensitivity analysis of datacenter cooling efficiency relative to IT rack density

Strengths

• Includes reference OpenFOAM configuration files for simulation reproducibility
• Covers four distinct geometric parameters: length, height, width, and rack count
• Provides normalized simulation data for standardized model training

Limitations

• Assumes static maximum flow rate and rack exit temperature, which may not reflect dynamic real-world workloads
• Restricted to hot-aisle geometry, excluding other datacenter cooling architectures like cold-aisle containment

Provenance

Source

NVIDIA

Collection Method

synthetic

Freshness

Last updated January 2026.