首先使用Anaconda创建一个 3.5的环境

然后使用pip install -i https://pypi.tuna.tsinghua.edu.cn/simple tf-nightly-gpu 安装gpu版本，版本号是tf_nightly_gpu-1.14.1.dev20190525，
上面这种安装方式可以避免包冲突。

保证 显卡驱动版本不低于425，然后安装cuda 10.0，不能是10.1，因为tensorflow-gpu会去找cufft64_100.dll,如果安装了其他版本，就找不到这个。

再安装cudnn，7.5版本就行了。 下载地址在常用库安装那一篇有。
安装cudnn只需要把三个文件复制到cuda的目录下就行

以上操作结束，就可以运行了。

完整环境：

Package              Version
-------------------- --------------------
absl-py              0.7.1
astor                0.8.0
certifi              2018.8.24
gast                 0.2.2
google-pasta         0.1.6
grpcio               1.21.1
h5py                 2.9.0
Keras-Applications   1.0.7
Keras-Preprocessing  1.0.9
Markdown             3.1.1
numpy                1.16.3
pip                  10.0.1
protobuf             3.7.1
setuptools           40.2.0
six                  1.12.0
tb-nightly           1.14.0a20190525
termcolor            1.1.0
tf-estimator-nightly 1.14.0.dev2019052401
tf-nightly-gpu       1.14.1.dev20190525
Werkzeug             0.15.4
wheel                0.31.1
wincertstore         0.2
wrapt                1.11.1

升级包：

pip install --upgrade  -i https://pypi.tuna.tsinghua.edu.cn/simple tf-nightly-gpu

升级python环境：
conda install -c anaconda python

安装好了可以运行以下代码，如果结果差不多，就是没问题

import tensorflow as tf
import numpy as np
import datetime

starttime = datetime.datetime.now()
# 制造数据
train_X = np.linspace(-1,1,100)
train_Y = 2*train_X+np.random.randn(*train_X.shape)*0.3
# y =2x+b


# 训练模型

train_epochs = 200
display_step =4 #展示模型参数
def moving_average(a,w=10):
    if len(a)<w:
        return a[:]
    return [val if i<w else sum(a[(i-w):i])/w for i,val in enumerate(a)]

# config = tf.ConfigProto(log_device_placement = True,allow_soft_placement=True)
# config.gpu_options.allow_growth = True
with tf.Session() as sess:
    # with tf.device("/gpu:0"):
        # 创建模型
        X = tf.placeholder("float")
        Y = tf.placeholder("float")
        W = tf.Variable(tf.random_normal([1]), name="weight")
        b = tf.Variable(tf.zeros([1]), name='bias')
        z = tf.multiply(X, W) + b

        cost = tf.reduce_mean(tf.square(Y - z))  # 损失函数
        learning_rate = 0.01  # 学习率，越小精度越高，速度越慢
        optimizer = tf.train.GradientDescentOptimizer(learning_rate).minimize(cost)  # 梯度下降算法

        pltdata = {'batchsize': [], "loss": []}  # 存放批次值和损失值
        sess.run(tf.global_variables_initializer())

        # 向模型填充数据
        for epoch in range(train_epochs):
            for (x, y) in zip(train_X, train_Y):
                sess.run(optimizer, feed_dict={X: x, Y: y})
                if epoch % display_step == 0:
                    loss = sess.run(cost, feed_dict={X: train_X, Y: train_Y})
                    print("Epoch", epoch + 1, "cost", loss, "W=", sess.run(W), "b=", sess.run(b))
                    if not (loss == 'NA'):
                        pltdata['batchsize'].append(epoch)
                        pltdata['loss'].append(loss)

        print('完成了。')
        print('cost=',sess.run(cost,feed_dict={X:train_X,Y:train_Y}),'W=',sess.run(W),'b=',sess.run(b))
        # plt.show()
        #使用模型预测
        print(sess.run(z,feed_dict={X:0.2}))
endtime = datetime.datetime.now()

print((endtime - starttime).seconds)

运行结果

C:\ProgramData\Anaconda3\envs\deepg\python.exe C:/projects/p520/p1.py
WARNING: Logging before flag parsing goes to stderr.
W0526 10:26:26.659331  2308 deprecation_wrapper.py:119] From C:/projects/p520/p1.py:23: The name tf.Session is deprecated. Please use tf.compat.v1.Session instead.

2019-05-26 10:26:26.668002: I tensorflow/stream_executor/platform/default/dso_loader.cc:42] Successfully opened dynamic library nvcuda.dll
2019-05-26 10:26:26.752551: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1640] Found device 0 with properties: 
name: GeForce RTX 2080 Ti major: 7 minor: 5 memoryClockRate(GHz): 1.755
pciBusID: 0000:01:00.0
2019-05-26 10:26:26.752719: I tensorflow/stream_executor/platform/default/dlopen_checker_stub.cc:25] GPU libraries are statically linked, skip dlopen check.
2019-05-26 10:26:26.753032: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1763] Adding visible gpu devices: 0
2019-05-26 10:26:26.753378: I tensorflow/core/platform/cpu_feature_guard.cc:142] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2
2019-05-26 10:26:26.754949: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1640] Found device 0 with properties: 
name: GeForce RTX 2080 Ti major: 7 minor: 5 memoryClockRate(GHz): 1.755
pciBusID: 0000:01:00.0
2019-05-26 10:26:26.755085: I tensorflow/stream_executor/platform/default/dlopen_checker_stub.cc:25] GPU libraries are statically linked, skip dlopen check.
2019-05-26 10:26:26.755354: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1763] Adding visible gpu devices: 0
2019-05-26 10:26:27.334236: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1181] Device interconnect StreamExecutor with strength 1 edge matrix:
2019-05-26 10:26:27.334359: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1187]      0 
2019-05-26 10:26:27.334435: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1200] 0:   N 
2019-05-26 10:26:27.334946: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1326] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 8694 MB memory) -> physical GPU (device: 0, name: GeForce RTX 2080 Ti, pci bus id: 0000:01:00.0, compute capability: 7.5)
W0526 10:26:27.336520  2308 deprecation_wrapper.py:119] From C:/projects/p520/p1.py:26: The name tf.placeholder is deprecated. Please use tf.compat.v1.placeholder instead.

W0526 10:26:27.337517  2308 deprecation_wrapper.py:119] From C:/projects/p520/p1.py:28: The name tf.random_normal is deprecated. Please use tf.random.normal instead.

W0526 10:26:27.345496  2308 deprecation_wrapper.py:119] From C:/projects/p520/p1.py:34: The name tf.train.GradientDescentOptimizer is deprecated. Please use tf.compat.v1.train.GradientDescentOptimizer instead.

W0526 10:26:27.371427  2308 deprecation_wrapper.py:119] From C:/projects/p520/p1.py:37: The name tf.global_variables_initializer is deprecated. Please use tf.compat.v1.global_variables_initializer instead.

Epoch 1 cost 0.7211578 W= [0.63478076] b= [-0.03276369]
Epoch 1 cost 0.70252275 W= [0.6578704] b= [-0.05632939]
Epoch 1 cost 0.68547904 W= [0.6809186] b= [-0.08034799]
Epoch 1 cost 0.6711159 W= [0.70232224] b= [-0.10313256]
Epoch 1 cost 0.6598311 W= [0.72094744] b= [-0.12339516]
Epoch 1 cost 0.6521096 W= [0.7350165] b= [-0.13904506]
Epoch 1 cost 0.64300734 W= [0.75364625] b= [-0.1602444]
Epoch 1 cost 0.63379073 W= [0.776071] b= [-0.18636262]
Epoch 1 cost 0.6308198 W= [0.78463596] b= [-0.19657867]
Epoch 1 cost 0.6242468 W= [0.8084542] b= [-0.22568986]
Epoch 1 cost 0.62169486 W= [0.8217175] b= [-0.24231094]
Epoch 1 cost 0.62036306 W= [0.83193773] b= [-0.25545126]
Epoch 1 cost 0.6200135 W= [0.8361488] b= [-0.26100987]
。。。


Epoch 153 cost 0.09066088 W= [1.9745536] b= [0.00509782]
Epoch 153 cost 0.09065532 W= [1.9748045] b= [0.00535933]
Epoch 153 cost 0.09061071 W= [1.9772507] b= [0.007856]
Epoch 153 cost 0.0905913 W= [1.978773] b= [0.00937827]
Epoch 157 cost 0.09057157 W= [1.984189] b= [0.00396216]
Epoch 157 cost 0.0905696 W= [1.9820896] b= [0.00610489]
Epoch 157 cost 0.09056988 W= [1.9819486] b= [0.00625191]
Epoch 157 cost 0.09056974 W= [1.9820172] b= [0.00617894]
Epoch 157 cost 0.090572976 W= [1.9810274] b= [0.00725581]
Epoch 157 cost 0.09061596 W= [1.9770781] b= [0.01164888]

Epoch 165 cost 0.09081806 W= [1.987512] b= [-0.00785136]
Epoch 165 cost 0.09068552 W= [1.9866278] b= [-0.00324412]
Epoch 165 cost 0.0906076 W= [1.9859699] b= [0.00058711]
Epoch 165 cost 0.09073182 W= [1.9868168] b= [-0.0050025]
Epoch 165 cost 0.09069343 W= [1.9866265] b= [-0.00355392]
Epoch 165 cost 0.09054514 W= [1.9855735] b= [0.00592261]
Epoch 165 cost 0.090541765 W= [1.9855192] b= [0.00652042]
Epoch 165 cost 0.0905363 W= [1.9854151] b= [0.00799263]
Epoch 165 cost 0.090542965 W= [1.9854988] b= [0.00633499]
Epoch 165 cost 0.09055714 W= [1.9855549] b= [0.00448342]
Epoch 165 cost 0.09053434 W= [1.9855] b= [0.00992332]
Epoch 165 cost 0.09056041 W= [1.9855462] b= [0.01449925]
Epoch 165 cost 0.09053523 W= [1.9853697] b= [0.00867187]


Epoch 173 cost 0.090757035 W= [1.9922602] b= [-0.00646905]
Epoch 173 cost 0.09082645 W= [1.9939458] b= [-0.00858147]
Epoch 173 cost 0.090811804 W= [1.9936193] b= [-0.00816173]
Epoch 173 cost 0.09062753 W= [1.988243] b= [-0.0010649]
Epoch 173 cost 0.0906094 W= [1.9874208] b= [5.0043625e-05]
Epoch 173 cost 0.09056159 W= [1.9824991] b= [0.00691267]
Epoch 173 cost 0.09057003 W= [1.9846733] b= [0.00379327]
Epoch 173 cost 0.09056097 W= [1.9826927] b= [0.00671967]

Epoch 173 cost 0.0905957 W= [1.9855791] b= [0.00144337]
Epoch 173 cost 0.09056565 W= [1.9849229] b= [0.00404223]
Epoch 173 cost 0.09060622 W= [1.9856844] b= [0.00076465]
Epoch 173 cost 0.09081806 W= [1.987512] b= [-0.00785136]
Epoch 173 cost 0.09068552 W= [1.9866278] b= [-0.00324412]
Epoch 173 cost 0.0906076 W= [1.9859699] b= [0.00058711]
Epoch 173 cost 0.09073182 W= [1.9868168] b= [-0.0050025]
Epoch 173 cost 0.09069343 W= [1.9866265] b= [-0.00355392]
Epoch 173 cost 0.09054514 W= [1.9855735] b= [0.00592261]
Epoch 173 cost 0.090541765 W= [1.9855192] b= [0.00652042]
Epoch 173 cost 0.0905363 W= [1.9854151]