Skip to content

Latest commit

 

History

History
112 lines (91 loc) · 3.34 KB

Install-CUDA-Drivers.md

File metadata and controls

112 lines (91 loc) · 3.34 KB

Install CUDA drivers

Following instructions assume you have a CUDA compatible GPU with at least 8GB VRAM (GTX1070 or better) as part of hardware.

Ubuntu 20.04

The installation of PyTorch GPU in Ubuntu 20.04 can be summarized in the following points, • Install CUDA by installing nvidia-cuda-toolkit.
• Install the cuDNN version compatible with CUDA.
• Export CUDA environment variables.

Installing CUDA

First open a terminal and run

$ sudo apt install nvidia-cuda-toolkit

which directly installs the latest version of CUDA in Ubuntu. After installing CUDA, run

$ nvcc -V

You will get an output similar to the following to verify if you had a successful installation,

nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2019 NVIDIA Corporation
Built on Sun_Jul_28_19:07:16_PDT_2019
Cuda compilation tools, release 10.1, V10.1.243

Note the CUDA version above release 10.1

Installing CUDNN

After above step, visit - https://developer.nvidia.com/rdp/cudnn-download - and download the CUDNN package that matches your CUDA version which is highlighted above. Once downloaded run

$ tar -xvzf cudnn-10.1-linux-x64-v7.6.5.32.tgz

Note that the package name might vary in your case. Now move the extracted packages like so

$ sudo cp cuda/include/cudnn.h /usr/lib/cuda/include/
$ sudo cp cuda/lib64/libcudnn* /usr/lib/cuda/lib64/

Set the file permissions of cuDNN,

$ sudo chmod a+r /usr/lib/cuda/include/cudnn.h 
$ sudo chmod a+r /usr/lib/cuda/lib64/libcudnn*

Export CUDA environment variables

The CUDA environment variables are needed by PyTorch for GPU support. To set them, we need to append them to ~/.bashrc file by running,

$ echo 'export LD_LIBRARY_PATH=/usr/lib/cuda/lib64:$LD_LIBRARY_PATH' >> ~/.bashrc
$ echo 'export LD_LIBRARY_PATH=/usr/lib/cuda/include:$LD_LIBRARY_PATH' >> ~/.bashrc

Load the exported environment variables by running,

$ source ~/.bashrc

Finally we can check if everything went fine by running

$ nvidia-smi

Windows

The installation of PyTorch GPU in Windows 10 can be summarized in the following points, • Install CUDA
• Install the cuDNN version compatible with CUDA.
• Export CUDA environment variables.

Install CUDA Toolkit

Visit https://developer.nvidia.com/cuda-downloads and download the cuda-toolkit from here

Installing CUDNN

After above step, visit - https://developer.nvidia.com/rdp/cudnn-download - and download the CUDNN package that matches your CUDA version. Once downloaded unzip the cuDNN package.

cudnn-x.x-windows-x64-vx.x.x.x.zip

Copy the following files into the CUDA Toolkit directory.

  1. Copy
<installpath>\cuda\bin\cudnn*.dll 
to 
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\vx.x\bin
  1. Copy
<installpath>\cuda\include\cudnn*.h 
to 
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\vx.x\include.
  1. Copy
<installpath>\cuda\lib\x64\cudnn*.lib
to 
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\vx.x\lib\x64.

Adding CUDA_PATH to environment variables

Variable Name: CUDA_PATH
Variable Value: C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\vx.x=

Finally we can check if everything went fine by running from a new command prompt.

$ nvidia-smi