Support for GPU

[beniz]

GPU offer up to 50 times faster linear algebra computations, including singular value decomposition.

There are three possible paths:

• Using a GPU software package that brings that connects to Eigen3 matrix library. Here, the simplest package seems to be https://github.com/rudaoshi/gpumatrix My current experiments show that gpumatrix types are missing some key operations of Eigen3, so it probably could only help with the sampling step of ES algorithms;
• Running the singular value decomposition on the GPU, using NVIDIA's tools for dense matrices: http://www.culatools.com/
• When the objective function is prohibitive, and that it is possible and appropriate to do so, running it on GPU might help.

[cmastalli]

is there any news about the libcmaes with GPU processing?

[beniz]

Hi, not that I know of. However, with the definite rise of GPUs, I believe this is still a good idea! In the past I did use the GPU to run the objective function though.