首先可以說明一點的是,在ubuntu16.04環境下CUDA8.0和CUDA7.0可以共存。筆者自己先安裝的是CUDA8.0,然後再安裝的CUDA7.0，安裝完CUDA8.0發現安裝CUDA7.0一直不成功，後面發現是16.04系統預設版本的GCC、g++版本過高，因此後面又裝了gcc-4.8,g++-4.8，即安裝了多版本的gcc和g++。

1.安裝cuda-8.0

sudo dpkg -i cuda-repo-ubuntu1404-8-0-local-ga2_8.0.61-1_amd64.deb
sudo apt-get update
sudo apt-get install cuda

2.配置環境變數

export LD_LIBRARY_PATH="$LD_LIBRARY_PATH:/usr/local/cuda/lib64"
export CUDA_HOME=/usr/local/cuda
export PATH="$CUDA_HOME/bin:$PATH"

3.安裝cuda-8.0對應的cudnn v5.1

tar xvzf cudnn-8.0-linux-x64-v5.1.tgz
mv cuda cudnn-8.0-v5.1
sudo cp cudnn-8.0-v5.1/include/cudnn.h /usr/local/cuda-8.0/include
sudo cp cudnn-8.0-v5.1/lib64/libcudnn*
 
 /usr/local/cuda-8.0/lib64

到此cuda-8.0及其對應的cudnn就安裝結束了。

4.安裝多版本CUDA

我先安裝的是CUDA8.0 後面再安裝CUDA7.0
由於當前系統的GCC G++都是5.4.0 對於安裝Cuda7.0來說版本太高，因此需要安裝低版本的GCC G++ 4.8
原本 /usr/bin目錄下
gcc軟連線到系統預設的GCC-5.4.0
gcc -> gcc-5
g++連結到系統預設的g++-5.4.0
g++ -> g++-5
首先安裝 gcc-4.8 g++-4.8

sudo apt-get install -y gcc-4.8
sudo apt-get install -y g++-4.8
#現在需要把軟連線到GCC-4.8 g++-4.8
sudo rm gcc 
sudo rm g++
sudo ln -s gcc-4.8 gcc
sudo ln -s g++-4.8 g++

接著安裝cuda-7.0和對應的cudnn

./cuda_7.0.28_linux.run
# 給cuda7.0安裝對應的cudnn cudnn-7.0-v4.0
tar xvzf cudnn-7.0-linux-x64-v4.0-prod.tgz 
mv cuda cudnn-7.0-v4.0
sudo cp cudnn-7.0-v4.0/include/cudnn.h   /usr/local/cuda-7.0/include/
sudo cp cudnn-7.0-v4.0/lib64/libcudnn*  /usr/local/cuda-7.0/lib64/

安裝過程中注意事項：
Do you accept the previously read EULA? (accept/decline/quit): accept
You are attempting to install on an unsupported configuration. Do you wish to continue? ((y)es/(n)o) [ default is no ]: y
Install NVIDIA Accelerated Graphics Driver for Linux-x86_64 346.46? ((y)es/(n)o/(q)uit): n
Do you want to install the OpenGL libraries? ((y)es/(n)o/(q)uit) [ default is yes ]: n
Install the CUDA 7.0 Toolkit? ((y)es/(n)o/(q)uit): y
Enter Toolkit Location [ default is /usr/local/cuda-7.0 ]:
/usr/local/cuda-7.0 is not writable.
Do you wish to run the installation with ‘sudo’? ((y)es/(n)o): y
Please enter your password:
Do you want to install a symbolic link at /usr/local/cuda? ((y)es/(n)o/(q)uit): n
Install the CUDA 7.0 Samples? ((y)es/(n)o/(q)uit): y
Enter CUDA Samples Location [ default is /home/wty ]:
Installing the CUDA Toolkit in /usr/local/cuda-7.0 …
Installing the CUDA Samples in /home/wty …
Copying samples to /home/wty/NVIDIA_CUDA-7.0_Samples now…
Finished copying samples.

You should pay attention carefully to the prompts offered you during install.
1. Select no when prompted to install the driver from the CUDA 7.5 runfile installer. Keep your CUDA 8 installed driver. It will work with CUDA 7.5
2. Select no when prompted to update the symlink for /usr/local/cuda (unless you want to switch the default config from CUDA 8 to CUDA 7.5)

When you want to actually switch your machine behavior from CUDA 8 to CUDA 7.0, you will need to modify the PATH and LD_LIBRARY_PATH env variables accordingly.

If you previous CUDA 8 install was via runfile installer, this should work fine. I use this method regularly. If you previous CUDA 8 install was via package manager method, I think this should still work but I don’t typically do that.

it’s probably also possible to do side-by-side install using only package manager methods, but I don’t have a set of instructions handy for that.