暗影精灵跑深度学习，环境安装：ubuntu16.04+GTX1050TI+cuda10.1+cudnn+tensorflow1.13

心已赠人 2022-01-21 03:07 573阅读 0赞

最近在用暗影精灵跑深度学习(基于tensorflow），随着数据量增多，CPU已经明显太慢，效率太低，所以把系统环境重新安装了一遍，搭建GPU环境。

机器平台: I7+1050TI+UBUNTU16.04

# 1.安装驱动： #

参考之前的一篇博文:[暗影精灵2代：ubuntu16.04 安装NVIDIA GTX1050TI显卡驱动][2_ubuntu16.04 _NVIDIA GTX1050TI]

这个可能会很慢，受不了的话可以到官网下载对应驱动来装(切勿使用CUDA默认的驱动，会导致无限登录问题)

官网链接：[https://www.nvidia.com/Download/index.aspx?lang=en-us][https_www.nvidia.com_Download_index.aspx_lang_en-us]

由于最新版CUDA10.1要求驱动高于418.39，参见：[https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html][https_docs.nvidia.com_cuda_cuda-toolkit-release-notes_index.html]

安装4.30无法启动界面，换成418.56，驱动正常，亦符合CUDA版本要求

# 2. CUDA #

官网最新版：[https://developer.nvidia.com/cuda-downloads?target\_os=Linux&target\_arch=x86\_64&target\_distro=Ubuntu&target\_version=1604&target\_type=runfilelocal][https_developer.nvidia.com_cuda-downloads_target_os_Linux_target_arch_x86_64_target_distro_Ubuntu_target_version_1604_target_type_runfilelocal]

按提示安装即可，注意去掉驱动安装项;

安装完成后，记得添加环境变量：

export PATH=/usr/local/cuda-10.1/bin:$PATH  
export LD\_LIBRARY\_PATH=/usr/local/cuda-10.1/lib64:$LD\_LIBRARY\_PATH

我安装的是最新10.1版本

测试：

进入~/NVIDIA\_CUDA-10.1\_Samples/1\_Utilities/deviceQuery

sudo make

./`deviceQuery`

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2x1dGVyZXNh_size_16_color_FFFFFF_t_70][]

# 3.安装cudnn #

官网：[https://developer.nvidia.com/rdp/cudnn-download][https_developer.nvidia.com_rdp_cudnn-download]

解压得到cuda目录，将其内容拷贝到对应的 `/usr/local/cuda-10.1`文件夹下即可

# 4.anconda #

参考之前的一篇博文：[Ubuntu16.04 安装anaconda3支持python3][Ubuntu16.04 _anaconda3_python3]

清华的源阵亡了，我直接官网下的；

我装的python3.7版，同样记得添加环境变量；

# 5.安装tensorflow-gpu #

conda create -n tf\_gpu

conda install tensorflow-gpu

![20190607205247477.png][]

[2_ubuntu16.04 _NVIDIA GTX1050TI]: https://blog.csdn.net/luteresa/article/details/79555356
[https_www.nvidia.com_Download_index.aspx_lang_en-us]: https://www.nvidia.com/Download/index.aspx?lang=en-us
[https_docs.nvidia.com_cuda_cuda-toolkit-release-notes_index.html]: https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html
[https_developer.nvidia.com_cuda-downloads_target_os_Linux_target_arch_x86_64_target_distro_Ubuntu_target_version_1604_target_type_runfilelocal]: https://developer.nvidia.com/cuda-downloads?target_os=Linux&target_arch=x86_64&target_distro=Ubuntu&target_version=1604&target_type=runfilelocal
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2x1dGVyZXNh_size_16_color_FFFFFF_t_70]: /images/20220121/e2421fcb0a4c46398bc59a60c8733f9b.png
[https_developer.nvidia.com_rdp_cudnn-download]: https://developer.nvidia.com/rdp/cudnn-download
[Ubuntu16.04 _anaconda3_python3]: https://blog.csdn.net/luteresa/article/details/79819476
[20190607205247477.png]: /images/20220121/fde24de2e411493a9085e4452b490b09.png