The NVIDIA cuSOLVER library provides a collection of dense and sparse direct linear solvers and Eigen solvers, delivering significant acceleration for applications in computer vision, computational fluid dynamics (CFD), computational chemistry, and linear optimization. The cuSOLVER library is included in both the NVIDIA HPC SDK and the CUDA Toolkit, making it accessible for a wide range of high-performance computing (HPC) applications.

Features

• cusolverDN: Key LAPACK dense solvers that are 3-6x faster than MKL, including dense Cholesky, LU, SVD, and QR solvers. These solvers are ideal for optimization, computer vision, and CFD applications.
• cusolverSP: Sparse direct solvers and symmetric & generalized symmetric eigensolvers, suitable for applications such as Newton's method and chemical kinetics.
• cusolverRF: Sparse refactorization solver, useful for applications in chemistry, ordinary differential equations (ODEs), and circuit simulation.
• cuSOLVERMp: A high-performance, distributed-memory, GPU-accelerated library for solving dense linear systems and eigenvalue problems. Available as a standalone download and included in the NVIDIA HPC SDK.
• Performance Optimization: Leverages DMMA Tensor Cores automatically, with DGX A100 being over 2x faster than DGX-2 despite having half the number of GPUs, thanks to A100 and third-generation NVLINK and NVSWITCH.

Benefits

• High Performance: Provides significant acceleration for dense and sparse linear solvers and Eigen solvers, enhancing computational efficiency.
• Scalability: Supports multi-node and multi-GPU operations, making it suitable for large-scale HPC applications.
• Versatility: Applicable to a wide range of scientific and engineering applications, including computer vision, CFD, computational chemistry, and linear optimization.
• Efficiency: Optimized for NVIDIA GPUs, leveraging advanced features like DMMA Tensor Cores for improved performance.