Compute / HPC

ScientificML

A PDE evaluation dataset containing 32 physical systems, each with four representations: source code, natural-language description, governing equation, and numerical trajectory, with both correct and corrupted variants.

ScientificML

A code perturbation dataset with 256 PDE solver implementations comprising 64 base programs and 4 lexical perturbation variants each, designed for evaluating LLM understanding of physics simulation code.

Multiphysics / PDE - 16 Domains

Unified benchmark spanning 16 physical domains, dubbed the 'ImageNet moment of physical ML', with PyTorch API, pretrained baselines and HF streaming.

Data Analysis / Parallel Computing

Distributed DataFrame library, a drop-in pandas replacement that auto-parallelizes across cores/clusters (Dask/Ray/Intel) for large-scale simulation data analysis.

Numerical Solver Library / HPC

High-performance scientific linear/nonlinear solver library for large sparse systems - the underlying engine of top-tier CFD and multiphysics simulation.

Solver / Training Framework

NVIDIA's physics-ML training framework (formerly Modulus) with PINN, neural operators and physics-constrained training - commercial-friendly, industry-first choice.

ScientificML

Warp-Geo provides a GPU-accelerated differentiable framework for reconstructing signed distance fields and evaluating surface normals from complex 3D point clouds, enabling gradient-based shape optimization.

PDE Solving

DeepXDE v1.15.0 released with L-LAAF activation function support for JAX backend and Apple MPS GPU acceleration for PyTorch backend, improving deep learning-based PDE solving performance.

Inference Optimization

TensorRT Edge-LLM achieves significant inference acceleration on Jetson AGX Thor edge platform, substantially improving AI agent deployment performance at the edge.

CFD

NVIDIA Earth-2 leverages AI atmospheric models integrated with observation data to provide high-resolution weather forecasts and decision support for weather-sensitive industries.

AI Inference Optimization

NVIDIA Groq 3 LPX leverages deterministic execution technology to enhance inference efficiency on the Vera Rubin platform, enabling high-interactivity and power-efficient computing for AI factory workloads.

Multibody

NVIDIA tutorial demonstrating GPU acceleration techniques for ROS 2 computational graphs using Isaac ROS and AI Agent integration to speed up robotics workloads.

ScientificML

Proposes a collaborative framework of AI virtual labs under a citation-based influence economy to automate neural operator architecture search and training evaluation.

Electromagnetics

Proposes a 3D FDTD framework using Quantized Tensor Trains that reduces memory and computational costs from conventional volumetric scaling to logarithmic scaling, enabling high-resolution electromagnetic scattering simulations with limited resources.

Framework Paper / Neural Operator

NVIDIA's PINN framework built on neural operators (FNO/DeepONet/AFNO) for end-to-end differentiable physics simulation, with multi-GPU training and deployment.

Must-Read Benchmark Paper

Proposes a unified benchmark across 16 physical domains, arguing for cross-domain pretraining gains - a must-read for physics foundation models.