Ubuntu16 GTX750 installation CUDA9.0, Pytorch, Anaconda Guide
Pre-installation warning
Do not use Ubuntu18!
Do not use Ubuntu18!
Do not use Ubuntu18!
Be sure to reinstall Ubuntu16
Driver installation NVDIA
This is only one way, once and for all
https://www.linuxidc.com/Linux/2019-02/157171.htm
Disable nouveau driver
sudo gedit /etc/modprobe.d/blacklist.conf
add
blacklist nouveau
Options nouveau modeset = 0
sudo-initramfs -u Update
with reboot lsmod | grep nouveau, if there is no output specification is disabled success.
https://blog.csdn.net/weixin_43820996/article/details/100676292
Installation CUDA9.0 (adaptation GTX750)
Download runfile (local) installation package
Do not rush to install, be sure NVIDIA will be given an official instruction manual carefully look
Directly open a command line in the graphical interface, according to the official installation tutorial
https://blog.csdn.net/QLULIBIN/article/details/78714596
Installation cudnn
This library installation is very simple, only need to copy files to a specified folder, according to the official tutorial
https://blog.csdn.net/wsc12358/article/details/81273884
Anaconda installation
Anaconda advantage is that you can create a virtual Python environment, which means you do not need to change the original Python Ubuntu, how to install themselves in Anaconda unload all, the domestic source added after installation
Creation Environment
conda create -n mypytorch python=3.7
Activation environment
source activate mypytorch
Install the corresponding version Pytorch
This step is more error-prone, mainly due to conda installation Pytorch always download times out, it is necessary to download a good local file installation
Reference Pytorch official document https://pytorch.org/get-started/previous-versions/
CUDA9.0 support the latest version of Pytorch1.1
Linux and Windows CUDA 9.0
conda install pytorch==1.1.0 torchvision==0.3.0 cudatoolkit=9.0 -c pytorch
This command should not be installed on, because pytorch download time out
Download a local file
To Tsinghua mirror source download the corresponding version pytorch
The file name is pytorch-1.1.0-py3.7_cuda9.0.176_cudnn7.5.1_0.tar.bz2, torchvision-0.3.0-py37_cu9.0.176_1.tar.bz2
conda local installation
conda install --use-local path/pytorch-1.1.0-py3.7_cuda9.0.176_cudnn7.5.1_0.tar.bz2
Execution Pytorch official command again
At this time, conda automatically skips installed package
conda install pytorch == 1.1.0 torchvision == 0.3.0 cudatoolkit = 9.0 (-c pytorch may be removed, do not remember)
Because failure is easy to install, so we can guarantee unlimited Anaconda fail
https://www.jianshu.com/p/76bbf1e5ff70 this blog a big help, thanks!