tensorflow with gpu 环境配置

一、准备工做

1.1 确保GPU驱动已经安装

lspci | grep -i nvidia 经过此命令能够查看GPU信息，测试机已经安装GPU驱动

nvidia-smi 能够查看英伟达显卡信息

1.2 确保gcc安装

能够经过gcc -v 查看，若是没有安装须要安装

1.3 确保安装open-ssh

若是没有安装能够经过 yum install openssh-server 安装

1.4 确保安装kernel

 sudo yum install kernel-devel-$(uname -r) kernel-headers-$(uname -r)

二、安装CUDA工具包

To use TensorFlow with NVIDIA GPUs, the first step is to install the  CUDA Toolkit.

备注：测试机选用的是 CUDA 8.0，不要使用9.x 有坑

三、安装GPU加速器cuDNN

安装完CUDA就能够安装  cuDNN .

备注：这个地方要选用与CUDA版本匹配的加速器；测试机选用的是： Download cuDNN v6.0 (April 27, 2017), for CUDA 8.0，安装步骤以下图所示：

四、安装或更新pip（若是有须要的话，若是已经有了，能够选择性跳过）

TensorFlow itself can be installed using the pip package manager. First, make sure that your system has pip installed and updated:

$ sudo apt-get install python-pip python-dev

$ pip install --upgrade pip

五、安装TensorFlow

Run the following command to install the TensorFlow Python package using pip:

$ pip install --upgrade tensorflow-gpu

六、测试安装是否成功

To test the installation, open an interactive Python shell and import the TensorFlow module:

 

(shan) root@ VM-241-228-ubuntu:/data/shan# python

Python 3.5.3 (default, Jun 23 2017, 16:12:41)

[GCC 5.4.0 20160609] on linux

Type "help", "copyright", "credits" or "license" for more information.

>>> import tensorflow as tf

>>> sess = tf.Session()

2017-07-12 19:24:14.030098: I tensorflow/stream_executor/cuda/ cuda_gpu_executor.cc:893] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero

2017-07-12 19:24:14.030833: I tensorflow/core/common_runtime/gpu/ gpu_device.cc:940] Found device 0 with properties:

name: Tesla M40 24GB

major: 5 minor: 2 memoryClockRate (GHz) 1.112

pciBusID 0000:00:06.0

Total memory: 22.40GiB

Free memory: 22.29GiB

2017-07-12 19:24:14.030855: I tensorflow/core/common_runtime/gpu/ gpu_device.cc:961] DMA: 0

2017-07-12 19:24:14.030867: I tensorflow/core/common_runtime/gpu/ gpu_device.cc:971] 0: Y

2017-07-12 19:24:14.030882: I tensorflow/core/common_runtime/gpu/ gpu_device.cc:1030] Creating TensorFlow device (/gpu:0) -> (device: 0, name: Tesla M40 24GB, pci bus id: 0000:00:06.0)

>>> hello_world = tf.constant("Hello, TensorFlow!")

>>> print (sess.run(hello_world))

b'Hello, TensorFlow!'

>>> print (sess.run(tf.constant(123)*tf.constant(456)))

56088

>>>

参考连接

• 官网安装文档
• cuDNN 安装
• CUDA安装