Hi Lev, On Sun, 3 Apr 2011 19:12:36 -0400, Lev Givon <l...@columbia.edu> wrote: > When I attempt to initialize PyCUDA immediately before forking a > process via the multiprocessing module and then try to initialize > PyCUDA within the new process, it appears that the forked process > turns into a zombie. Is this behavior expected? To observe this, > uncomment the drv.init() in the main body of the attached script. > > I'm using PyCUDA 0.94.2, Python 2.6.5, pyzmq 2.1.1, and zeromq 2.1.4.
I asked this a long while back on the Nvidia forums: http://forums.nvidia.com/index.php?showtopic=83219&st=0&p=471611&hl=fork&fromsearch=1&#entry471611 My results were essentially the same as yours. It seems that you can only fork before you initialize CUDA. (Bummer, I know.) If you find out anything different, please let me (or better, the list) know. Andreas
pgphD7tmHd4Qd.pgp
Description: PGP signature
_______________________________________________ PyCUDA mailing list PyCUDA@tiker.net http://lists.tiker.net/listinfo/pycuda
