pycuda anyone using it?


I’ve seen the home page of pycuda:

and it looks quite interesting! Since programming under python is WAY easier than C…

is there any mayor drawback in using python?

I supposed that there would be a substantial efficiency lost, but they claim that there’s almost none. So maybe you lose flexibility? control over the hardware?

I would like to start programming, but being a mathematician rather than a programmer, this would help me a lot to concentrate on the algorithm rather than on its implementation…

Side note: If you know MATLAB, you can just install the Accelereyes Jacket program, which adds CUDA acceleration to MATLAB. I figured since you’re a mathematician, you probably have some experience with it…

yes thank you!

i can’t wait for my geforce to arrive to port my algoritms! the fact is that i may need stand alone applications…

I’m fairly sure that you can export/compile MATLAB programs to native code (i.e. an exe or DLL), and I do know that MATLAB exposes a COM server, so you can make calls to MATLAB from another program (e.g. to run the algorithm ‘behind the scenes’)


profquail is correct… you can invoke and run MATLAB code/scripts (including Jacket code/scripts) from external applications without having the user of the external application know anything about MATLAB (except for the fact that it must be installed on the machine). In fact, I recently implemented this for an open source visualization tool, Slicer3 (see…rExampleModule).

But, in the end, we realize that many people will want to be able to put Jacket code directly into end products/applications. So we have as a priority on our roadmap the creation of an avenue for people to statically compile Jacket’ed CUDA code into an application.