Available Versions Feature list Licenses

We present GPU-accelerated iterative solvers for large systems of sparse linear equations. The performance of the library was tested on matrices from the University of Florida Sparse Matrix Collection and compared against CUSPARSE and CUSP.

In average SpeedIT 2.0 outperforms CUSPARSE 4.0 and CUSP 0.2.0 by a factor of x2 and x4, respectively.

See our results for details on performance.

» SpeedIT 2.0

• CG and BCGSTAB
• Single & double precision
• Sparse Matrix Vector Multiplication
• Diagonal preconditioner
• Support for complex numbers

» Requirements for SpeedIT and OpenFOAM Plugin

• Linux 32-bit/64-bit
• CUDA Toolkit 4.0
• GPU card compatible with CUDA standard with compute capability 2.0 or higher
• gcc 4.4.3

» Roadmap

• AMG, AINV preconditioners
• Support for multi-gpu