AmgX is a high-performance, state-of-the-art library designed to accelerate core solver technology on NVIDIA GPUs. It offers up to 10x acceleration for the computationally intense linear solver portion of simulations, making it especially well-suited for implicit unstructured methods. AmgX includes a flexible solver composition system that allows users to easily construct complex nested solvers and preconditioners. The library provides optimized methods for massive parallelism and is accessible through a simple C API that abstracts the parallelism and GPU implementation.

Features

• Flexible Configuration: Allows for nested solvers, smoothers, and preconditioners.
• Algebraic Multigrid Methods: Includes Ruge-Steuben and un-smoothed aggregation algebraic multigrid methods.
• Krylov Methods: Supports PCG, GMRES, BiCGStab, and flexible variants.
• Smoothers: Offers Block-Jacobi, Gauss-Seidel, incomplete LU, Polynomial, and dense LU smoothers.
• Support for Various Systems: Handles scalar or coupled block systems.
• Parallel Computing Support: Includes MPI and OpenMP support.
• High-Level C API: Provides a flexible and simple high-level C API for easy integration.

Benefits

• Significant Performance Gains: Delivers up to 10x acceleration for linear solvers, enhancing computational efficiency.
• Versatility: Applicable to a wide range of domains, including CFD, aerospace, automotive design, heavy industry, energy, and more.
• Ease of Use: Simple C API abstracts the complexity of parallelism and GPU implementation, making it accessible for developers.
• Scalability: Optimized for massive parallelism, supporting large-scale simulations and computations.
• Flexibility: Allows users to construct complex nested solvers and preconditioners, providing tailored solutions for specific applications.