CentOS 7 builds Linux GPU server

1. CUDA Toolkit Installation

To Https://developer.nvidia.com/cuda-gpus query GPU-supported CUDA versions:

To Https://developer.nvidia.com/cuda-downloads, according to the operating system choose to download the appropriate CUDA toolkit version, download is a. run file, the download is completed with the root user directly run the file installation.

After the installation is finished. Run:

Nvidia-smi

If the GPU status information is listed, it indicates that the installation was successful:

2. Installation of CUDNN

TensorFlow the acceleration of the neural network through the CUDNN Library implementation, so first go to https://developer.nvidia.com/rdp/cudnn-archive, according to cuda version download the corresponding version of CUDNN, is also a. run file that runs directly after the download is complete.

3. Installation of TensorFlow

In order to avoid issues such as version conflicts during installation, it is recommended that you install Anoconda first. To https://www.anaconda.com/download/#linux下载后, run the. sh file installation.

Then install TensorFlow using the following command:

Conda create-n tensorflow python=2.7source activate Tensorflowexport tf_binary_url=https://storage.googleapis.com/ Tensorflow/linux/gpu/tensorflow-0.11.0rc1-cp27-none-linux_x86_64.whlpip Install--ignore-installed--upgrade $TF _ Binary_url

Enter in turn:

SOURCE Activate Tensorflowpythonimport TensorFlow as Tfimport pandas as pdtf.__version__

If no error is found, the installation is successful:

CentOS 7 builds Linux GPU server