Any benchmark for CHOLMOD on GPU

Hi,

I need this information regarding a funding proposal I am resubmitting.

I have been asked to justify proposed purchase of GPU and report some benchmark against CPU for my application. Most time consuming part of my algorithm involving sparse matrix is CHOLMOD. I am planning to use the one on Prof. Tim Davis' website which is already optimized for GPU:

http://www.cise.ufl.edu/research/sparse/cholmod/

Any benchmark for either his CHOLMOD algorithm or any CHOLMOD algorithm would be appreciated.

I am expecting my sparse matrix size to be 7000 x 11000.

IF no benchmark can be suggested, can someone suggest me how should I convince reviewers by doing back-of-the-paper calculation.

Regards, DK

[Louis]

I am also interested in this.

I am currently trying to build Cholmod with GPU under Linux

There is no support in Cholmod for GPU support with the example Mex files created in the cholmod/matlab folder

Louis

[ngimel]

http://on-demand.gputechconf.com/gtc/2014/presentations/S4201-gpu-acceleration-sparse-matrix-factorization-cholmod.pdf

For your matrix size, you are unlikely to see much benefit, though