NVIDIA cuSPARSE is a GPU-accelerated library that provides basic linear algebra subroutines for sparse matrix computations. It is widely used by engineers and scientists working on applications in machine learning, AI, computational fluid dynamics (CFD), seismic exploration, and computational sciences. The cuSPARSE 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

• Support for Multiple Sparse Matrix Formats: Includes dense, COO, CSR, CSC, and Blocked CSR formats.
• Comprehensive Sparse Routines: Covers sparse vector x dense vector operations, sparse matrix x dense vector operations, and sparse matrix x dense matrix operations.
• Sparse Matrix Addition and Multiplication: Provides routines for sparse matrix x sparse matrix addition and multiplication.
• High-Performance APIs: Offers generic high-performance APIs for sparse-dense vector multiplication (SpVV), sparse matrix-dense vector multiplication (SpMV), and sparse matrix-dense matrix multiplication (SpMM).
• Preconditioners: Includes ILU0 and IC0 preconditioners for improving the convergence of iterative solvers.
• cuSPARSELt: Provides structured sparsity support via Sparse Tensor Core in Ampere and later architectures for General Matrix Multiplications (GEMMs), with options for pruning and compression of sparse matrices, activation functions, bias vectors, and output scaling for AI and deep learning use cases.

Benefits

• Enhanced Performance: Delivers significant performance improvements on NVIDIA GPUs, with SpMM performance 30-150X faster than CPU-only alternatives.
• Versatility: Applicable to a wide range of scientific and engineering applications, including machine learning, AI, CFD, and seismic exploration.
• Scalability: Supports large-scale HPC applications with multi-GPU and multi-node capabilities.
• Efficiency: Optimized for NVIDIA GPUs, leveraging advanced features like Sparse Tensor Cores for improved performance.
• Reliability: Provides high-quality, tested routines for sparse matrix computations, ensuring accuracy and stability.