Tensorflow使用GPU訓練

確認顯卡驅動正確安裝：

(notebook) [wuhf@aps ~]$ nvidia-smi
Thu Aug 20 18:07:33 2020
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 430.50       Driver Version: 430.50       CUDA Version: 10.1     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  GeForce GTX 1080    Off  | 00000000:82:00.0 Off |                  N/A |
| 27%   29C    P8     6W / 180W |    113MiB /  8119MiB |      0%      Default |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes:                                                       GPU Memory |
|  GPU       PID   Type   Process name                             Usage      |
|=============================================================================|
|    0     37575      C   ...uhf/miniconda3/envs/notebook/bin/python   103MiB |
+-----------------------------------------------------------------------------+

CUDA版本和Tensorflow版本有對應關系，TF2.0可以使用CUDA 10.1，安裝TF2.0版本，查看conda 源中的TF ：

(notebook) [wuhf@aps ~]$ conda search tensorflow |grep 2.0.0
tensorflow                     2.0.0 eigen_py27hec4e49e_0  pkgs/main
tensorflow                     2.0.0 eigen_py36ha83d16c_0  pkgs/main
tensorflow                     2.0.0 eigen_py37hce6be7f_0  pkgs/main
tensorflow                     2.0.0 gpu_py27hb041a2f_0  pkgs/main
tensorflow                     2.0.0 gpu_py36h6b29c10_0  pkgs/main
tensorflow                     2.0.0 gpu_py37h768510d_0  pkgs/main
tensorflow                     2.0.0 mkl_py27h68eb67f_0  pkgs/main
tensorflow                     2.0.0 mkl_py36hef7ec59_0  pkgs/main
tensorflow                     2.0.0 mkl_py37h66b46cc_0  pkgs/mai

一定要安裝 gpu的build，指定build安裝方法：

conda install {project}={version}={build}

執行命令：

conda install tensorflow=2.0.0=gpu_py36h6b29c10_0

然后來執行python代碼測試TF是否正常：

import tensorflow as tf
tf.test.is_gpu_available()
tf.test.gpu_device_name()

輸出：

Type "help", "copyright", "credits" or "license" for more information.
>>> import tensorflow as tf
>>> tf.test.is_gpu_available()
True
>>> tf.test.gpu_device_name()
2020-08-20 18:16:13.857330: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1618] Found device 0 with properties:
'/device:GPU:0'
>>>

如果安裝很慢，可使用國內源，在用戶目錄下新建.condarc文件，內容如下：

channels:
  - defaults
show_channel_urls: true
channel_alias: https://mirrors.tuna.tsinghua.edu.cn/anaconda
default_channels:
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/free
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/r
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/pro
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/msys2
custom_channels:
  conda-forge: https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud
  msys2: https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud
  bioconda: https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud
  menpo: https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud
  pytorch: https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud
  simpleitk: https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud

使用pip安裝

不管是gpu還是cpu版本的tf，安裝文件都會復制到{python}/lib/python3.7/site-packages/tensorflow 這樣后安裝就會覆蓋新安裝的。

關鍵點總結：

1. tensorflow一定要安裝帶gpu的build
2. tensorflow-gpu這個不用安裝
3. tensorflow的版本要和cuda版本匹配，可以去官網查看對應關系

資料：
Tensorflow與CUDA之間的關系： https://tensorflow.google.cn/install/gpu