cuSPARSE
Sparse Linear Algebra
The NVIDIA CUDA Sparse Matrix library (cuSPARSE) provides GPU-accelerated basic linear algebra subroutines for sparse matrices that perform up to 5x faster than CPU-only alternatives. You can use the flexible C and C++ interface to sparse routines, pre-conditioners, optimized precision computation (double, single, half) and data storage formats to develop applications that deliver the highest performance with least effort.
cuSPARSE is widely used by engineers and scientists working on applications such as machine learning and natural language processing, computational fluid dynamics, seismic exploration and computational sciences.
Download Now Explore what’s new in the latest release...High Performance
The CUDA software platform is highly optimized for execution on NVIDIA GPUs and performs 2X-5X faster than CPU-only alternatives.
Key Features
- Supports dense, COO, CSR, CSC, ELL/HYB and Blocked CSR sparse matrix formats
- High-performance sparse-dense matrix multiplies (SpMM)
- Optimized sparse matrix-vector (SpMV) for unstructured matrices with highly variable nnzs per row
- Level 1 routines for sparse vector x dense vector operations
- Level 2 routines for sparse matrix x dense vector operations
- Level 3 routines for sparse matrix x multiple dense vectors (tall matrix)
- Routines for sparse matrix by sparse matrix addition and multiplication
- Conversion routines that allow conversion between different matrix formats
- Sparse Triangular Solve
- Tri-diagonal solver
- Incomplete factorization preconditioners ilu0 and ic0
Product Resources
Availability
The cuSPARSE library is freely available as part of the CUDA Toolkit.
For more information on cuSPARSE:
- Source code examples demonstrating how to use the cuSPARSE library:
- White Paper: "Incomplete-LU and Cholesky Preconditioned Iterative Methods Using CUSPARSE and CUBLAS
- GPU Accelerated Solution of Sparse Linear Systems