Incubator-mxnet: cublas error when using like mx.nd.***

[tornadomeet]

hello,all
when using mx.nd function, sometimes it output the error like this:
./dmlc-core/include/dmlc/logging.h:208: [17:19:40] ./mshadow/mshadow/./stream_gpu-inl.h:98: Check failed: (err) == (CUBLAS_STATUS_SUCCESS) Create cublas handle failed terminate called after throwing an instance of 'dmlc::Error' terminate called recursively
it is strange for me that all the example and test is correct.
so the simple code which outputs error above is(test.py here):

import mxnet as mx
import numpy as np
import logging
logger = logging.getLogger()
logger.setLevel(logging.INFO)
ctx = mx.gpu(0)
if __name__ == "__main__":
    mx.nd.zeros((3L, 1L, 5L, 5L), ctx)

and run in shell: python -u test.py, it will error, but when run line by line in python terminal, no erroe there, why?
ths~

[piiswrong]

there are issues with singleton destruction in mxnet when your program shutdown before mxnet is fully initialized. Just add a time.sleep(5) at the end and it should be find.

[tornadomeet]

@piiswrong thanks, i'll try to do this,

[tengpeng]

I confirm the solution worked. If not, try a larger number, like 35 seconds.

[FCInter]

I met with the similar problem and the solution provided here solves my problem.
Thank you.

[eric-haibin-lin]

or just put a mx.nd.waitall()