Setup GPU specific packages on Windows

  • Article

This section outlines the packages you need to setup in order for CNTK to leverage NVIDIA GPUs.

Checking your GPU compatibility

You need a CUDA-compatible graphic card to use CNTK GPU capabilities. You can check whether your card is CUDA-compatible here and here (for older cards). Your GPU card Compute Capability (CC) must be 3.0 or more.

In the following steps we will install the NVidia development tools required to build the Microsoft Cognitive Toolkit as well as NVidia support libraries. As the last step (after you installed all aforementioned NVidia tools!), you should check that you have the latest graphic card driver installed.

Make sure the directory C:\Program Files\NVIDIA Corporation\NVSMI exists in your system.

  • Quick installation check: If you followed the instruction above and used the same paths, the command dir C:\Program Files\NVIDIA Corporation\NVSMI\nvml.dll will succeed.

Latest GPU card driver

Install the latest driver for your GPU card:

  • Select your card and download the driver pack from this download location
  • Run the driver installation procedure

NVIDIA CUDA 9.0

Download and install the NVIDIA CUDA 9.0 Toolkit:

Make sure that the following CUDA environment variables are set to the correct path (the NVIDIA Cuda installer will create these for you). Default installation paths are assumed:

CUDA_PATH="C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.0"
CUDA_PATH_V9_0="C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.0"
  • Quick installation check: If you followed the instruction above and used the same paths, the command dir C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.0\bin\cudart64_90.dll will succeed.

cuDNN

Install NVIDIA CUDA Deep Neural Network library also known as cuDNN in the version NVIDIA: cuDNN v7.0 for CUDA 9.0 from this link. This version is suitable for Windows 8.1, Windows 10, as well as Windows Server 2012 R2 and later.

  • Extract the archive to a folder on your local disk, e.g. to C:\local\cudnn-9.0-v7.0\

  • Quick installation check: If you followed the instruction above and used the same paths, the command dir C:\local\cudnn-9.0-v7.0\cuda\bin\cudnn64_7.dll will succeed.

CUB

Important

If you are installing CNTK for Python, you may skip this step.

Important

Install NVIDIA CUB using the exact version specified below. This is necessary because it is expected by the CNTK build configuration program.

  • Download NVIDIA CUB v.1.7.4 from this download link

  • Extract the archive to a folder on your local disk (we assume c:\local\cub-1.7.4).

  • Quick installation check. If you followed the instruction above and used the same paths, this command dir C:\local\cub-1.7.4\cub\cub.cuh will succeed.