NVIDIA SimNet^{TM}: an AI-accelerated multi-physics simulation framework

TL;DR

SimNet tackles the computational bottleneck of traditional PDE solvers in complex, multi-physics environments by delivering an AI-driven, parameterized neural solver that can perform forward, inverse, and data-assimilation tasks in a single trained model. The framework combines geometry modules (CSG/TG), PDE modules (including zero-equation turbulence and exact/integral continuity), and advanced network architectures (Fourier features, SiReNs) to enable rapid, GPU-accelerated solutions for real-world geometries. Key innovations include SDF-based loss weighting, integral continuity planes, and full geometry/physics parameterization, which together enable data-free forward solves and efficient design optimization across large design spaces. Real-world use cases demonstrate strong agreement with OpenFOAM and commercial solvers and reveal substantial speedups for parameterized design exploration, validating SimNet’s potential for accelerating engineering design, optimization, and inverse problems in industry and academia.

Abstract

We present SimNet, an AI-driven multi-physics simulation framework, to accelerate simulations across a wide range of disciplines in science and engineering. Compared to traditional numerical solvers, SimNet addresses a wide range of use cases - coupled forward simulations without any training data, inverse and data assimilation problems. SimNet offers fast turnaround time by enabling parameterized system representation that solves for multiple configurations simultaneously, as opposed to the traditional solvers that solve for one configuration at a time. SimNet is integrated with parameterized constructive solid geometry as well as STL modules to generate point clouds. Furthermore, it is customizable with APIs that enable user extensions to geometry, physics and network architecture. It has advanced network architectures that are optimized for high-performance GPU computing, and offers scalable performance for multi-GPU and multi-Node implementation with accelerated linear algebra as well as FP32, FP64 and TF32 computations. In this paper we review the neural network solver methodology, the SimNet architecture, and the various features that are needed for effective solution of the PDEs. We present real-world use cases that range from challenging forward multi-physics simulations with turbulence and complex 3D geometries, to industrial design optimization and inverse problems that are not addressed efficiently by the traditional solvers. Extensive comparisons of SimNet results with open source and commercial solvers show good correlation.

Figures (15)

• Figure 1: Design optimization of an FPGA heat sink using SimNet. The center and side fin heights are the two design variables, with a linear transition for the height of the intermediate fins.
• Figure 2: Four major areas in computational science and engineering addressed by SimNet.
• Figure 3: A schematic of the structure of a neural network solver. The inputs to the network are the spatial coordinates of a point cloud, realizations of time (if applicable), and realizations from the parametric space (if applicable). The inputs are mapped to the quantities of interest via a fully-connected network with nonlinear activation functions. To train this network, a loss function is considered which consists of the derivatives of the output w.r.t inputs (computed using automatic differentiation), and initial and boundary condition information.
• Figure 4: SimNet structure.
• Figure 5: FPGA heat sink example. (a) The FPGA heat sink geometry; (b) the simulation domain. The blue plane represents the plane of symmetry.