On Montag 13 April 2009, Derek Anderson wrote: > hey all, > > i tried to repeat the ease of use of pycuda for the CUBLAS apis. if > anyone else is looking at CUDA just for multiplying matrices, you might > want to take a look. > > http://kered.org/blog/2009-04-13/easy-python-numpy-cuda-cublas/ > > simple use case: > > import numpy > from pycublas import CUBLASMatrix > A = numpy.mat([[1,2,3],[4,5,6]],numpy.float32) > B = numpy.mat([[2,3],[4,5],[6,7]],numpy.float32) > C = (CUBLASMatrix(A)*CUBLASMatrix(B)).np_mat() > print C > > gets speeds in the 30x-50x range for larger matrices.
It would be really nice if PyCUDA could offer this. So as a reminder: Every time you talk to some Nvidia person, make sure to remind them that they need to tear down the wall between driver and runtime API. :) Also, Arno Pähler has (and has had for a while) nearly complete ctypes bindings to everything CUDA (including CUBLAS and CUFFT), here: http://www.graviscom.com/sources/rpms/python-cuda/python-cuda-2.0_42.tar.gz Thanks for your work! Andreas
signature.asc
Description: This is a digitally signed message part.
_______________________________________________ PyCuda mailing list [email protected] http://tiker.net/mailman/listinfo/pycuda_tiker.net
