I bought Colab pro to train a model written with MXNet. In the standard version of Colab with GPU, everything is working fine. But in the pro version, a simple array allocation on GPU is running forever: mx.nd.array([1, 2, 3], ctx=mx.gpu(0))
.
Versions:
- cuda 10.1 installed via
!apt-get -y install cuda-10-1
- mxnet with Cuda support 10.1 installed via
%pip install mxnet-cu101==1.5.0
The image below basically says that the execution of the cell started at 13:50 (6 minutes ago). 6 minutes to allocate an array of 3 values? Why is this happening?