1、Windows版
准备
干净的系统，没有安装过Python，有的话就卸载了。
另外我的系统安装了VS2015 VS2017（这里我不知道是不是必备的）。
现在TensorFlow和cuda以及cuDNN品名升级，所以这里采用了几乎是最新版的了（2018年11月19日）
- Anaconda——清华tuna下载
- 显卡驱动——点我去英伟达官网自行下载对应驱动
- cuda9.0安装包——点我去百度云下载
- cuDNN7.x安装包——点我去百度云下载
安装
1、安装Anaconda
这里省略。注意一点，安装的选项加入path，都勾选。
2、安装显卡驱动
默认安装。
3、安装cuda9.0
默认安装。
4、安装cuDNN 7.x
将压缩包解压，放在C:\ProgramData\NVIDIA GPU Computing Toolkit\v9.0这个目录下。
然后将目录C:\ProgramData\NVIDIA GPU Computing Toolkit\v9.0\bin添加到环境变量PATH里。
验证
1、启动Anaconda Prompt
输入
1 conda env list
显示只有一个base或者root的环境。表示只有一个环境。
2、修改Anaconda的软件源
执行
1 conda config --add channels https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/free/
2 conda config --add channels https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/
3 conda config --set show_channel_urls yes
表示将anaconda的软件下载源修改成清华Tuna的了。
3、创建用于TensorFlow的Python环境
conda create -n tf-gpu-py3.5 python=3.5
例子:
D:\Users\zyb>conda create -n tf-gpu-py3.5 python=3.5
Solving environment: done## Package Plan ## environment location: C:\anaconda35\envs\tf-gpu-py3.5 added / updated specs:
- python=3.5The following NEW packages will be INSTALLED: certifi: 2018.8.24-py35_1001 https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/conda-forge
pip: 18.0-py35_1001 https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/conda-forge
python: 3.5.5-he025d50_2 https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/conda-forge
setuptools: 40.4.3-py35_0 https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/conda-forge
vc: 14.1-h21ff451_1 https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/peterjc123
vs2017_runtime: 15.4.27004.2010-1 https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/peterjc123
wheel: 0.32.0-py35_1000 https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/conda-forge
wincertstore: 0.2-py35_1002 https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/conda-forgeProceed ([y]/n)? y
Preparing transaction: done
Verifying transaction: done
Executing transaction: done
#
# To activate this environment, use
#
# $ conda activate tf-gpu-py3.5
#
# To deactivate an active environment, use
#
# $ conda deactivate
4、激活刚刚创建的环境
conda activate tf-gpu-py3.5
5、安装TensorFlow GPU版
conda install tensorflow-gpu
6、代码验证
启动python
输入如下代码
import tensorflow as tf
查看是否报错。
如果报错，就使用conda install 包名（比如numpy）
如果不报错，接着执行
1 a = tf.constant([1.0,2.0,3.0,4.0,5.0,6.0],shape=[2,3],name='a')
2 b = tf.constant([1.0,2.0,3.0,4.0,5.0,6.0],shape=[3,2],name='b')
3 c = tf.matmul(a,b)
4 sess = tf.Session(config=tf.ConfigProto(log_device_placement=True))
5 #这步结束之后，会出现一个警告：
6 #Device mapping: no known devices.
7 #2018-11-19 22:18:15.899459: I T:\src\github\tensorflow\tensorflow\core\common_runtime\direct_session.cc:288] Device mapping:
8 #不用管，执行下一步
9 print(sess.run(c))
10 #输出如下：
11 MatMul: (MatMul): /job:localhost/replica:0/task:0/device:CPU:0
12 2018-11-19 22:18:23.059234: I T:\src\github\tensorflow\tensorflow\core\common_runtime\placer.cc:935] MatMul: (MatMul)/job:localhost/replica:0/task:0/device:CPU:0
13 a: (Const): /job:localhost/replica:0/task:0/device:CPU:0
14 2018-11-19 22:18:23.064109: I T:\src\github\tensorflow\tensorflow\core\common_runtime\placer.cc:935] a: (Const)/job:localhost/replica:0/task:0/device:CPU:0
15 b: (Const): /job:localhost/replica:0/task:0/device:CPU:0
16 2018-11-19 22:18:23.069134: I T:\src\github\tensorflow\tensorflow\core\common_runtime\placer.cc:935] b: (Const)/job:localhost/replica:0/task:0/device:CPU:0
17 [[22. 28.]
18 [49. 64.]]
验证成功。
2、Ubuntu下安装GPU版TensorFlow
准备
1、Anaconda-Linux版本的——去清华tuna自行下载
2、显卡驱动——去官网自行下载
3、cuda9.0——去官网自行下载Linux版本的
4、cuDNN7.x——去官网下载Linux版本的（需要注册并且join）
安装
1、Anaconda安装
这里需要注意，直接把软件安装在自己的家目录下即可。
2、安装显卡驱动
官网下载驱动，然后使用sudo安装。
安装的过程中，第一步需要你阅读安装协议。使用q退出。
3、安装cuda9.0
默认安装。
安装的过程中，第一步需要你阅读安装协议。使用q退出。
9.0有一个base安装包还有4个升级包。都是有序号的。
使用sudo chmod +x *.run给这5个文件加上可执行权限
然后一个个安装。
然后将安装完后的路径加入PATH环境变量。
1 export PATH=/usr/local/cuda-9.0/bin:/usr/local/cuda-9.0/lib64:$PATH
2 export LD_LIBRARY_PATH=/usr/local/cuda-9.0/lib64:$LD_LIBRARY_PATH
4、安装cuDNN
解压出来两个文件夹一个是include 一个是lib64
a.使用sudo将include里的cudnn.h文件复制到/usr/local/cuda-9.0/include 目录下
b.使用sudo将lib64里面的libcudnn.so.7.3.1 libcudnn_static.a两个文件复制到/usr/local/cuda-9.0/lib64 目录下。
c.做两个软连接，cd到/usr/local/cuda-9.0/lib64 目录下，执行：
1 sudo ln -s libcudnn.so.7.3.1 libcudnn.so
2 sudo ln -s libcudnn.so.7.3.1 libcudnn.so.7
验证
0、cuda验证
#进入样本目录
cd ~/home/NVIDIA_CUDA-9.0_Samples
#编译样本
make -j8
#进入生成可执行文件的目录
cd bin/x86_64/linux/release
#执行设备测试程序
./deviceQuery
#输出如下
./deviceQuery Starting... CUDA Device Query (Runtime API) version (CUDART static linking)Detected 1 CUDA Capable device(s)Device 0: "GeForce GTX 1070"
CUDA Driver Version / Runtime Version 10.0 / 9.0
CUDA Capability Major/Minor version number: 6.1
Total amount of global memory: 8116 MBytes (8510701568 bytes)
(15) Multiprocessors, (128) CUDA Cores/MP: 1920 CUDA Cores
GPU Max Clock rate: 1683 MHz (1.68 GHz)
Memory Clock rate: 4004 Mhz
Memory Bus Width: 256-bit
L2 Cache Size: 2097152 bytes
Maximum Texture Dimension Size (x,y,z) 1D=(131072), 2D=(131072, 65536), 3D=(16384, 16384, 16384)
Maximum Layered 1D Texture Size, (num) layers 1D=(32768), 2048 layers
Maximum Layered 2D Texture Size, (num) layers 2D=(32768, 32768), 2048 layers
Total amount of constant memory: 65536 bytes
Total amount of shared memory per block: 49152 bytes
Total number of registers available per block: 65536
Warp size: 32
Maximum number of threads per multiprocessor: 2048
Maximum number of threads per block: 1024
Max dimension size of a thread block (x,y,z): (1024, 1024, 64)
Max dimension size of a grid size (x,y,z): (2147483647, 65535, 65535)
Maximum memory pitch: 2147483647 bytes
Texture alignment: 512 bytes
Concurrent copy and kernel execution: Yes with 2 copy engine(s)
Run time limit on kernels: Yes
Integrated GPU sharing Host Memory: No
Support host page-locked memory mapping: Yes
Alignment requirement for Surfaces: Yes
Device has ECC support: Disabled
Device supports Unified Addressing (UVA): Yes
Supports Cooperative Kernel Launch: Yes
Supports MultiDevice Co-op Kernel Launch: Yes
Device PCI Domain ID / Bus ID / location ID: 0 / 1 / 0
Compute Mode:
< Default (multiple host threads can use ::cudaSetDevice() with device simultaneously) >deviceQuery, CUDA Driver = CUDART, CUDA Driver Version = 10.0, CUDA Runtime Version = 9.0, NumDevs = 1
Result = PASS
#看到PASS后执行带宽测试
./bandwidthTest
#输出如下：
[CUDA Bandwidth Test] - Starting...
Running on... Device 0: GeForce GTX 1070
Quick Mode Host to Device Bandwidth, 1 Device(s)
PINNED Memory Transfers
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 12758.2 Device to Host Bandwidth, 1 Device(s)
PINNED Memory Transfers
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 12867.2 Device to Device Bandwidth, 1 Device(s)
PINNED Memory Transfers
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 191582.5Result = PASSNOTE: The CUDA Samples are not meant for performance measurements. Results may vary when GPU Boost is enabled.
#看到PASS表示测试通过，如果FAIL，重启然后重新执行即可。
1、创建anaconda环境（和Windows一样）
1 conda create -n tf-gpu-py3.5 python=3.5
2 #
3 # To activate this environment, use
4 #
5 # $ conda activate tf-gpu-py3.5
6 #
7 # To deactivate an active environment, use
8 #
9 # $ conda deactivate
2、激活tf-gpu-py3.5
conda activate ty-py-3.5-cpu
3、安装tensorflow-gpu
conda install tensorflow-gpu
4、代码验证
(tf-gpu-py3.5) tf@lolita-ThinkStation-P318:~/anaconda3/envs$ python
Python 3.5.6 |Anaconda, Inc.| (default, Aug 26 2018, 21:41:56)
[GCC 7.3.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import tensorflow as tf
>>> a = tf.constant([1.0,2.0,3.0,4.0,5.0,6.0],shape=[2,3],name='a')
>>> b = tf.constant([1.0,2.0,3.0,4.0,5.0,6.0],shape=[3,2],name='b')
>>> c = tf.matmul(a,b)
>>> sess = tf.Session(config=tf.ConfigProto(log_device_placement=True))
2018-11-19 22:43:27.732910: I tensorflow/core/platform/cpu_feature_guard.cc:141] Your CPU supports instructions that this TensorFlow binary was not compiled to use: SSE4.1 SSE4.2 AVX AVX2 FMA
2018-11-19 22:43:27.824810: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:897] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2018-11-19 22:43:27.825419: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1405] Found device 0 with properties:
name: GeForce GTX 1070 major: 6 minor: 1 memoryClockRate(GHz): 1.683
pciBusID: 0000:01:00.0
totalMemory: 7.93GiB freeMemory: 7.64GiB
2018-11-19 22:43:27.825445: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1484] Adding visible gpu devices: 0
2018-11-19 22:43:27.995777: I tensorflow/core/common_runtime/gpu/gpu_device.cc:965] Device interconnect StreamExecutor with strength 1 edge matrix:
2018-11-19 22:43:27.995806: I tensorflow/core/common_runtime/gpu/gpu_device.cc:971] 0
2018-11-19 22:43:27.995826: I tensorflow/core/common_runtime/gpu/gpu_device.cc:984] 0: N
2018-11-19 22:43:27.996035: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1097] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 7377 MB memory) -> physical GPU (device: 0, name: GeForce GTX 1070, pci bus id: 0000:01:00.0, compute capability: 6.1)
Device mapping:
/job:localhost/replica:0/task:0/device:GPU:0 -> device: 0, name: GeForce GTX 1070, pci bus id: 0000:01:00.0, compute capability: 6.1
2018-11-19 22:43:28.026839: I tensorflow/core/common_runtime/direct_session.cc:288] Device mapping:
/job:localhost/replica:0/task:0/device:GPU:0 -> device: 0, name: GeForce GTX 1070, pci bus id: 0000:01:00.0, compute capability: 6.1>>> print(sess.run(c))
MatMul: (MatMul): /job:localhost/replica:0/task:0/device:GPU:0
2018-11-19 22:44:23.662448: I tensorflow/core/common_runtime/placer.cc:935] MatMul: (MatMul)/job:localhost/replica:0/task:0/device:GPU:0
a: (Const): /job:localhost/replica:0/task:0/device:GPU:0
2018-11-19 22:44:23.662561: I tensorflow/core/common_runtime/placer.cc:935] a: (Const)/job:localhost/replica:0/task:0/device:GPU:0
b: (Const): /job:localhost/replica:0/task:0/device:GPU:0
2018-11-19 22:44:23.662589: I tensorflow/core/common_runtime/placer.cc:935] b: (Const)/job:localhost/replica:0/task:0/device:GPU:0
[[22. 28.]
[49. 64.]]
验证完毕。
