Using CUDA?

seahen

I'm running RapidMiner on my gaming PC, to help inform the work I'm doing on LFMonitor (a real-time chat-message classifier). Is there a way to use CUDA (a GPGPU technology that my Nvidia GTX-480 card supports) to speed up processing?

Right now I'm running Compare ROCs with 4 different classifiers, looped (via Loop Parameters (Parallel) and Select Subprocess) over 4 different attribute sets, 2 of which are chosen by Optimize Solution (Evolutionary). It runs 65 threads with all but the main process parallelized, so I figure it should scale pretty well over CUDA.

On a related note, what's the optimum number of threads to have running? I have a Core i7 (4 physical x64 cores with Hyper-Threading) and 480 CUDA cores. I'm running Windows 7, but I know how to use Ubuntu and will gladly set up dual boot if it'll help. If I'm running too many threads, can I reduce the number without disabling parallelization on individual operators?

MariusHelf

Hi seahen,

to use CUDA in RapidMiner each operator to be used has to be ported to the CUDA architecture, which is a non-trivial and time consuming task. Currently some specialized operators are implemented for CUDA in scientific projects, but as far as I know they are not (yet) ready for public consumption. So currently I have to say that it is not yet possible to speed up RapidMiner with the use of CUDA.

To use your CPU cores optimally with the parallel extension RapidMiner should be configured to use as many cores as you have, and generally the greatest speed-up is achieved if you only parallelize the outermost loop.

Cheers,
Marius