Please download Windows and Linux version by clicking on "Browse All Files".
GPUmat allows standard MATLAB code to run on GPUs. The engine is written in C/C++ and based on NVIDIA CUDA.
Please contact firstname.lastname@example.org for any questions.
Unfortunately GPUmat was compiled for CUDA 5.0 and we basically stopped any support for other CUDA version because we don't have the resources to do it. But you can compile the source code if you want.
Instructions to compile the source code:
You need to download a SVN client and then from command line:
svn export http://svn.code.sf.net/p/gpumat/code/trunk ./GPUmat
the above command creates a folder ./GPUmat with the source code.
Then go to the 'doc' folder, open the GPUmat_Developer_Guide.pdf and
check the chapter "Source code compilation"
- Execute Matlab code on GPUs
This is awesome software! Using GPUmat on Nv670 I get speed-ups between x8 and x10 over a 4 core Phenome system for a method working on large dissimilarity matrices. I tried to combine with GNU/octave but failed despite of relocating mex files - I think octave support should be possible without major efforts. This would be fantastic. Anyway thanks so much to the developers for their great work! ... (PS.: would you like to support the sign() and bsxfun() functions in future?)