After installing the host first described Ubuntu18.04 do not install anything, including graphics driver, cuda graphics card installed inside the drive belt. Direct download cuda10.0 and the corresponding version of cudnn, mainly because at this stage the highest version TensorFlow support cuda10.0 seemingly, does not support cuda10.1 should be installed cuda10.1 run TensorFlow wrong for me. So the back and unload cuda10.1, installation cuda10.0.
First, install the CUDA
1. Download the
official website to download ( CUDA official website )
view download previous versions, go after selecting cuda10.0 download
and run the package in accordance with the option to download version
2. Installation
according to Description open a terminal run:
sudo sh cuda_10.0.130_410.48_linux.run
Terminal installation interface appears, the agreement appears, press the Enter key has been running (if you want to see can look down one by one, hee hee ......)
because I have already installed the card, so there I first graphics card chose n, others are y, if your ubuntu is clean, directly select all y. The next step is to wait for the installation was successful! After the write path environment variable is an input terminal (to be in the root directory):
sudo gedit .bashrc
If it is not the root directory, enter:
sudo gedit ~.bashrc
In the write path:
export CUDA_HOME=/usr/local/cuda
export PATH=$PATH:$CUDA_HOME/bin
export LD_LIBRARY_PATH=/usr/local/cuda-10.0/lib64${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}
Save and exit, restart .bashrc file:
source ~.bashrc
3. Test
the input terminal:
cd /usr/local/cuda/samples/1_Utilities/deviceQuery
sudo make
./deviceQuery
Appears in Figure indicates the installation was successful:
Second, install cudnn
1. Download
cudnn download the corresponding version (cudnn need to log in to download site, you can register by mail, then download the compressed file, as shown in Figure):
2. Install
directly with Ubuntu Right extract to the current file into the file, open a terminal and run:
1. sudo cp cuda/include/cudnn.h /usr/local/cuda/include/
2. sudo cp cuda/lib64/libcudnn* /usr/local/cuda/lib64/
3. sudo chmod a+r /usr/local/cuda/include/cudnn.h
4. sudo chmod a+r /usr/local/cuda/lib64/libcudnn*
View CUDNN version in a terminal:
cat /usr/local/cuda/include/cudnn.h | grep CUDNN_MAJOR -A 2
So far, cuda and cudnn been installed successfully!
Daniel middle reference to a number of blog, in this very grateful! ! !