ACTIVATE PhysicsNemo

Run NVIDIA PhysicsNemo examples on GPU clusters via the ACTIVATE platform.

Features

• Curated examples tuned for A30 24GB GPUs - works out-of-the-box
• Custom script mode for advanced users
• Pre-flight checks - GPU memory, Docker daemon, disk space
• Output collection - plots, metrics, and training summaries
• SLURM/PBS support via job_runner marketplace action

Curated Examples

Example	Method	Data	Runtime (A30)
Darcy Flow (FNO)	Fourier Neural Operator	Generated on-the-fly	~10 min
Lid-Driven Cavity (PINNs)	Physics-Informed NN	None (physics-only)	~15 min
Gray-Scott RNN	Recurrent NN	Generated	~20 min
Darcy Flow (PINO)	Physics-Informed NO	Generated + physics	~15 min
Vortex Shedding (MeshGraphNet)	Graph Neural Network	Included	~25 min

Usage

1. Select a GPU cluster resource
2. Choose a curated example or paste a custom script
3. (Optional) Override training epochs or batch size
4. Run the workflow

Outputs (plots, metrics) are collected to the outputs/ directory.

Custom Script Mode

Write any bash script to run inside the PhysicsNemo container. The working directory is /workspace. Example:

#!/bin/bash
git clone --depth 1 https://github.com/NVIDIA/PhysicsNemo.git /workspace/PhysicsNemo
cd /workspace/PhysicsNemo/examples/cfd/ldc_pinns
python -u train.py

Container

Default image: nvcr.io/nvidia/physicsnemo/physicsnemo:25.11

Supports both sudo Docker and rootless Docker modes.