获取 CUDA_HOME 环境路径 PYTORCH

Question

I have cuda installed via anaconda on my system which has 2 GPUs which is getting recognized by my python.

import torch
torch.cuda.is_available()
true

However when I try to run a model via its C API, I m getting following error:

~/anaconda3/lib/python3.6/site-packages/torch/utils/cpp_extension.py in _join_cuda_home(*paths)
722     '''
723     if CUDA_HOME is None:
--> 724         raise EnvironmentError('CUDA_HOME environment variable is not set. '
725                                'Please set it to your CUDA install root.')
726     return os.path.join(CUDA_HOME, *paths)

OSError: CUDA_HOME environment variable is not set. Please set it to your CUDA install root.

https://lfd.readthedocs.io/en/latest/install_gpu.html page gives instruction to set up CUDA_HOME path if cuda is installed via their method. Since I have installed cuda via anaconda I don't know which path to set.

I tried find method but it is returning me too many paths for cuda. Can somebody help me with the path for CUDA. Thanks in advance.

Answer 1

Solution to above issue!

As cuda installed through anaconda is not the entire package. Please install cuda drivers manually from Nvidia Website[ https://developer.nvidia.com/cuda-downloads ]

After installation of drivers, pytorch would be able to access the cuda path. You can test the cuda path using below sample code. Problem resolved!!!

CHECK INSTALLATION:

import os
print(os.environ.get('CUDA_PATH'))

OUTPUT:

C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.1