1

I tried to install mxnet with CUDA support and eventually succeeded. Although the installation finally worked, all my python applications are freezing whenever I try to actually use it. To be more precise, I'm trying to run the validation code:

import mxnet as mx
a = mx.nd.ones((2, 3), mx.gpu())
b = a * 2 + 1
b.asnumpy()

When running this, nothing happens and the cell in my jupyter notebook dies not terminate. Same in Visual Studio 2019. It works fine if I leave out the GPU part. I'm running python 3.8 through an anaconda installation, CUDA version 10.1 and mxnet-cu101 version 1.7.0. The notebook with this code is running in a virtual environment (mxnet-cu101 is of course also installed for that environment).

I already googled around a lot, but a lot of posts about this topic are pretty old and recommend switching between old versions of all these packages for the at that time current version of mxnet.

Edit: Although the python notebook just freezes, I get the following error in the anaconda prompt:

Upgrade advisory: this mxnet has been built against cuDNN lib version 7500, 
which is older than the oldest version tested by CI (7600).  Set 
MXNET_CUDNN_LIB_CHECKING=0 to quiet this warning.

My cuDNN version is 8.0.5.39, so i don't know how that relates to the warning message.

Tim Wiesner
  • 11
  • 3
  • you may wish to [verify](https://docs.nvidia.com/cuda/cuda-installation-guide-microsoft-windows/index.html#verify-installation) your CUDA install. You may wish to start at the beginning of that document to take note of the [requirements](https://docs.nvidia.com/cuda/cuda-installation-guide-microsoft-windows/index.html#system-requirements). For example you need a CUDA capable GPU. – Robert Crovella Jan 29 '21 at 16:51
  • @RobertCrovella Thanks for the tip, but I followed all the verification steps and everything works fine. My GPU is also CUDA capable. Does it make a difference that I'm running the notebook with this code in a virtual environment? – Tim Wiesner Jan 31 '21 at 14:22

0 Answers0