Unable to Connect Numba to CUDA Hardware on Windows 10 Machine

[Rehman Ali]

Hello Everyone,

I am a new Numba user, and I am trying to take advantage of the GPU on my Windows 10 system to speed up my numpy. However, I have not been unable to run the basic CUDA examples (see numbaexample.py attached). The first error I ran into was:

NvvmSupportError: libNvvm cannot be found. Do 'conda install cudatoolkit'

Since the system I am trying to install the cudatoolkit is on a networked system with a firewall I cannot remove, I was unable to run the simple package installer. So I downloaded the file from: https://anaconda.org/anaconda/cudatoolkit/files. Then I ran 'conda install win-64_cudatoolkit-7.5-0.tar.bz2'

After the install completed, I set the the following environment variables:

NUMBAPRO_LIBDEVICE as C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\nvvm\libdevice

NUMBAPRO_NVVM as C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\nvvm\lib\x64\nvvm.lib

I reran numbaexample.py. Unfortunately, I ran into another error which reads as follows:

C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\nvvm\lib\x64\nvvm.lib us either not a designed to run on Windows or it contains an error. Try installing the program again using the original installation media or contact your system administrator or the software vendor for support. Error status 0xc000012f

I am not sure what I am doing incorrectly here. I installed and reinstalled CUDA 8.0 then 7.5, and neither works. 

When I ran cuda.api.detect(), I found all 4 CUDA devices which are "[SUPPORTED]", so everything should have been fine.

Would someone please help me to solve this problem?

Thank You