CUSP is an open source C++ library of generic parallel algorithms for sparse linear algebra and graph computations on CUDA architecture GPUs. CUSP provides a flexible, highlevel interface for manipulating sparse matrices and solving sparse linear systems.

*NEW* CUDA 5 Math Library Performance Report Now Available
GPU: Tesla C2050 with CUDA 4.0 CPU: Core i7 950, Trlinos/ML v5.0 and MKL v10.3
The following example loads a matrix from disk, transparently converts the matrix to the highlyefficient HYB format, and transfers the matrix to the GPU device. The linear system A*x = b is them solved on the device using the Conjugate Gradient method. Additional code samples are available on the CUSP Google Code Project
#include<cusp/hyb_matrix.h> #include<cusp/io/matrix_market.h> #include<cusp/krylov/cg.h> int main(void) { // create an empty sparse matrix structure (HYB format) cusp::hyb_matrix<int, float,cusp::device_memory>A; // load a matrix stored in MatrixMarket format cusp::io::read_matrix_market_file(A,"5pt_10x10.mtx"); // allocate storage for solution (x) and right hand side (b) cusp::array1d<float, cusp::device_memory>x(A.num_rows,0); cusp::array1d<float, cusp::device_memory>b(A.num_rows,1); // solve the linear system A * x = b with the Conjugate Gradient method cusp::krylov::cg(A, x, b); return 0; } 
CUSP is available under the Apache 2.0 open source license on Google Code.
If you are looking for a productiontested sparse matrix library with a C API that supports sparse matrix operations and building blocks for sparse iterative solvers, learn more about cuSPARSE.