PhysicsNeMo-Mesh: GPU-Native Meshes and Simulation Cases for Physics AI
Abstract
PhysicsNeMo-Mesh is the mesh module of the open-source PhysicsNeMo library. It provides mesh processing similar to PyVista, but on the GPU, in batches, and inside PyTorch's automatic differentiation. Like xarray for unstructured meshes, it attaches semantics to data: fields on points, cells, and whole meshes have names and shapes, and a simulation case keeps the complete PDE problem specification, including boundary conditions and their prescribed values. It addresses three problems that plague machine learning for physical simulation. First, GPU utilization is bottlenecked by I/O, because samples are very large relative to the compute per forward pass. Second, because traditional preprocessing is expensive, it is necessarily done once, during curation, rather than differently for each model and workflow at the point of use. Finally, advanced equivariant and mesh-native architectures require information that curation ordinarily discards. Fast GPU mesh processing makes it practical to preprocess at the point of use: on one GB200 GPU, mesh operations run 6.2−−1,068× faster than VTK with 72 threads, and loading a whole case is 8.0−−88× faster than reading its VTK files (8.0× when both read the same fields). Overlapped with training, this preprocessing costs 2.1% of throughput for a 127 ms Transolver step. So one dataset serves many models and workflows: in PhysicsNeMo's public recipe, one stored DrivAerML dataset feeds Transolver, GeoTransolver, and GLOBE. PhysicsNeMo-Mesh combines the speed of an intermediate array store with the flexibility of an archival format such as VTK. When shapes and wall conditions both vary, per-boundary conditions cut a heat-conduction error from 87% to 6−−9%. And by keeping field ranks and named boundaries, it retains the semantic structure that next-generation architectures will need.
Related
- Numerical simulation of biological capsules based on interface-fitted polygonal meshes
- triangulax: Differentiable Simulations with Triangular Meshes across Soft Matter Physics
- diff-soup: GPU-native Differentiable Simulation and Inference for Large-scale Stiff Dynamical Systems
- JC-PIC: kinetic (PIC-MCC) simulation of lowtemperature plasmas with a documented library of one hundred test cases
- Passage of particles through matter and the effective straggling-function: High-fidelity accelerated simulation via Physics-Informed Machine Learning
BibTeX
@misc{sharpe2026physicsnemomesh,
title = {{PhysicsNeMo-Mesh: GPU-Native Meshes and Simulation Cases for Physics AI}},
author = {Peter D. Sharpe and Corey Adams and Negin Sobhani and Sanjay Choudhry},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.39072},
eprint = {2610.39072},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.39072},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - PhysicsNeMo-Mesh: GPU-Native Meshes and Simulation Cases for Physics AI
AU - Peter D. Sharpe
AU - Corey Adams
AU - Negin Sobhani
AU - Sanjay Choudhry
PY - 2026
DA - 2026/10/29
PB - arXiv
DO - 10.48550/arxiv.2610.39072
UR - https://arxiv.org/abs/2610.39072
AB - PhysicsNeMo-Mesh is the mesh module of the open-source PhysicsNeMo library. It provides mesh processing similar to PyVista, but on the GPU, in batches, and inside PyTorch's automatic differentiation. Like xarray for unstructured meshes, it attaches semantics to data: fields on points, cells, and whole meshes have names and shapes, and a simulation case keeps the complete PDE problem specification, including boundary conditions and their prescribed values. It addresses three problems that plague machine learning for physical simulation. First, GPU utilization is bottlenecked by I/O, because samples are very large relative to the compute per forward pass. Second, because traditional preprocessing is expensive, it is necessarily done once, during curation, rather than differently for each model and workflow at the point of use. Finally, advanced equivariant and mesh-native architectures require information that curation ordinarily discards. Fast GPU mesh processing makes it practical to preprocess at the point of use: on one GB200 GPU, mesh operations run 6.2−−1,068× faster than VTK with 72 threads, and loading a whole case is 8.0−−88× faster than reading its VTK files (8.0× when both read the same fields). Overlapped with training, this preprocessing costs 2.1% of throughput for a 127 ms Transolver step. So one dataset serves many models and workflows: in PhysicsNeMo's public recipe, one stored DrivAerML dataset feeds Transolver, GeoTransolver, and GLOBE. PhysicsNeMo-Mesh combines the speed of an intermediate array store with the flexibility of an archival format such as VTK. When shapes and wall conditions both vary, per-boundary conditions cut a heat-conduction error from 87% to 6−−9%. And by keeping field ranks and named boundaries, it retains the semantic structure that next-generation architectures will need.
ER -