You are here

GPGPU

HPCA GPGPU Software Page

Software request form

CULAPACK

Some of the dense linear algebra codes developed as part of the articles:

Solving dense linear systems on graphics processors

and

Exploiting the capabilities of modern GPUs for dense matrix computations

can be downloaded using the "Software request form" (follow link to the right or above). These include LAPACK-like code samples for the Cholesky factorization and the LU factorization with partial pivoting. Codes are provided under the BSD license.

FLAG/C - FLAG@lab

FLAG/C and FLAG@lab are respectively C and matlab application programming interfaces (APIs) to develop FLAME-like codes for dense linear algebra operations on GPUs. Refer to the FLAME webpage to learn more about this approach.

Read the FLAG/C Users Guide for more information.

SuperMatrix with multi-GPU support

SuperMatrix has recently added multi-GPU support, boosting performance of a dense Cholesky factorization code to nearly 650 GFLOPs for the Cholesky factorization on a Tesla S1070 system.

Some performance results can be found at the FLAME wiki and the approach is described in detail in the article

Solving Dense Linear Systems on Platforms with Multiple Hardware Accelerators

For more information about SuperMatrix and the multi-GPU support, please contact gquintan@icc.uji.es.