Research & Papers

WaterLily.jl scales to 1B cells with MPI and improved multigrid solver

Near-ideal scaling on CPUs/GPUs, weak efficiency above 96% inter-node

Deep Dive

WaterLily.jl, a scale-resolving incompressible flow solver written in pure Julia, now includes MPI-based parallelism. Strong-scalability tests show a near-ideal linear trend. Weak-scaling efficiency stays above 85% until node memory-concurrency contention dominates, and inter-node weak scalability exceeds 96% for grids up to 1 billion cells.

Key Points
  • MPI parallelism achieves near-ideal strong scaling and >96% inter-node weak scaling efficiency
  • Grid sizes up to 1 billion cells demonstrated with minimal overhead across distributed nodes
  • Improved geometric multigrid solver uses adaptive under-relaxed red-black Gauss-Seidel smoother

Why It Matters

Democratizes large-scale fluid dynamics simulation using a portable, pure-Julia solver with near-perfect scaling performance.

📬 Get the top 10 AI stories daily