Ubuntu16.04下安装多版本cuda和cudnn

ubuntu16.0.4 + TITAN X + anaconda3+ tensorflow-gpu(0.12.1)

电脑已经安装CUDA9+ cuDNN7.1，本次安装CUDA8.0.44 + cuDNN5.1

相关命令：

查看cuda版本： nvcc -V

查看位置： which nvcc

查看NVIDIA动态使用情况： watch -n 1 nvidia-smi

cuda 版本： cat /usr/local/cuda/version.txt

cudnn 版本： cat /usr/local/cuda/include/cudnn.h | grep CUDNN_MAJOR -A 2

NVIDIA 驱动版本： cat /proc/driver/nvidia/version

查看环境变量 : env

LD_DEBUG=all cat

卸载cuda ： sudo /usr/local/cuda-8.0/bin/uninstall_cuda_8.0.pl

卸载NVIDIA Driver ： sudo /usr/bin/nvidia-uninstall

多版本CUDA切换:

sudo rm -rf /usr/local/cuda

sudo ln -s /usr/local/cuda-8.0 /usr/local/cuda

sudo ln -s /usr/local/cuda-9.1 /usr/local/cuda

具有管理员权限的文件管理器, 比如移动文件夹 : sudo nautilus

加入-R 参数，将权限传递给子文件夹 : chmod -R 777 /home/mypackage

**********************************************************************************************************

GitHub上下了个程序，tensorflow-gpu=0.12，gpu下跑报错，应该是CUDA版本高了

E tensorflow/stream_executor/cuda/cuda_dnn.cc:378] Loaded runtime CuDNN library: 7103 (compatibility version 7100) but source was compiled with 5105 (compatibility version 5100).  If using a binary install, upgrade your CuDNN library to match.  If building from sources, make sure the library loaded at runtime matches a compatible version specified during compile configuration.
E tensorflow/stream_executor/cuda/cuda_dnn.cc:378] Loaded runtime CuDNN library: 7103 (compatibility version 7100) but source was compiled with 5105 (compatibility version 5100).  If using a binary install, upgrade your CuDNN library to match.  If building from sources, make sure the library loaded at runtime matches a compatible version specified during compile configuration.
F tensorflow/core/kernels/conv_ops.cc:532] Check failed: stream->parent()->GetConvolveAlgorithms(&algorithms) 
F tensorflow/core/kernels/conv_ops.cc:532] Check failed: stream->parent()->GetConvolveAlgorithms(&algorithms)

https://blog.csdn.net/tunhuzhuang1836/article/details/79545625 ：Ubuntu16.04下安装多版本cuda和cudnn

https://blog.csdn.net/Mr_KkTian/article/details/78632756 ：Ubuntu16.04下同时安装CUDA8.0和CUDA7.0

https://blog.csdn.net/maple2014/article/details/78574275 ：安装多版本 cuda ，多版本之间切换

https://blog.csdn.net/mumoDM/article/details/79462604 ：多版本CUDA问题

https://blog.csdn.net/liangyihuai/article/details/78688228 ：windows下tensorflow-gpu安装

0、 Tensorflow gpu 官方安装指南：

https://www.tensorflow.org/install/install_windows

下载CUDA并安装：
各个版本的CUDA ：https://developer.nvidia.com/cuda-toolkit-archive
下载CUDNN (要注册)
CUDNN库下载地址：https://developer.nvidia.com/cudnn

Installation Guide for Linux : cuda_8.0.44(官方安装说明) cuda_8.0.44

安装CUDA8.0和cuDNN5.1：

下载好后直接命令行解压然后复制 lib64 和 include 文件夹到 usr/local/cuda-8.0，命令如下：

# Installing from a Tar Filetar -zxvf 压缩文件名.tar.gz
sudo cp cuda/include/cudnn.h  /usr/local/cuda-8.0/include
sudo cp cuda/lib64/libcudnn*  /usr/local/cuda-8.0/lib64
sudo chmod a+r /usr/local/cuda-9.1/include/cudnn.h  /usr/local/cuda-8.0/lib64/libcudnn*

cuda版本切换

gedit ~/.bashrc #更改 ~/.bashrc 文件，添加两行

export PATH=/usr/local/cuda/bin${PATH:+:${PATH}}
export LD_LIBRARY_PATH=/usr/local/cuda/lib64${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}

下面的不行：

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

sudo /etc/profile #必须更改/etc/profile 文件, 而且更改后必须重启计算机才有效 (source /etc/profile 不能生效)

export PATH=/usr/local/cuda-9.1/bin:$PATH  
export LD_LIBRARY_PATH=/usr/local/cuda-9.1/lib64:$LD_LIBRARY_PATH

改为:

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

从cuda9.1切换到cuda8.0：sudo rm -rf /usr/local/cuda                       #删除之前创建的软链接
sudo ln -s /usr/local/cuda-8.0  /usr/local/cuda   #创建新 cuda 的软链接

从cuda8.0切换到cuda9.0：sudo rm -rf /usr/local/cuda                       #删除之前创建的软链接sudo ln -s /usr/local/cuda-9.1  /usr/local/cuda   #创建新 cuda 的软链接

可以用命令来查看cuda是否切换完成：

$ nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2017 NVIDIA Corporation
Built on Fri_Sep__1_21:08:03_CDT_2017
Cuda compilation tools, release 9.0, V9.0.176which nvcc  ：查看nvcc位置

CUDA8.0+cuDNN5.1 未完全安装：

Installing the CUDA Toolkit in /usr/local/cuda-8.0 ...
Installing the CUDA Samples in /home/human-machine ...
Copying samples to /home/human-machine/NVIDIA_CUDA-8.0_Samples now...
Finished copying samples.

===========
= Summary =
===========

Driver:   Not Selected
Toolkit:  Installed in /usr/local/cuda-8.0
Samples:  Installed in /home/human-machine

Please make sure that
 -   PATH includes /usr/local/cuda-8.0/bin
 -   LD_LIBRARY_PATH includes /usr/local/cuda-8.0/lib64, or, add /usr/local/cuda-8.0/lib64 to /etc/ld.so.conf and run ldconfig as root

To uninstall the CUDA Toolkit, run the uninstall script in /usr/local/cuda-8.0/bin

Please see CUDA_Installation_Guide_Linux.pdf in /usr/local/cuda-8.0/doc/pdf for detailed information on setting up CUDA.

***WARNING: Incomplete installation! This installation did not install the CUDA Driver. A driver of version at least 361.00 is required for CUDA 8.0 functionality to work.
To install the driver using this installer, run the following command, replacing <CudaInstaller> with the name of this run file:
    sudo <CudaInstaller>.run -silent -driver

Logfile is /tmp/cuda_install_1534.log

#编译并测试设备 deviceQuery：
切换到例子存放的路径，默认路径是 ~/NVIDIA_CUDA-7.5_Samples ，切换到相应路径
然后终端输入：$ make
运行编译生成的二进制文件
编译后的二进制文件, 默认存放在 ~/NVIDIA_CUDA-7.5_Samples/bin
切换路径：$ cd /NVIDIA_CUDA-8.0_Samples/bin/x86_64/linux/release
终端输入：$ ./deviceQuery

#编译并测试带宽 bandwidthTest：
cd ../bandwidthTest
sudo make
./bandwidthTest
如果这两个测试的最后结果都是Result = PASS，说明CUDA安装成功

CUDA8.0+cuDNN5.1报错，但tensorflow-gpu可以跑，tensorboard也可以用。

human-machine@humanmachine-System-Product-Name:~/NVIDIA_CUDA-8.0_Samples$ make
make[1]: Entering directory '/home/human-machine/NVIDIA_CUDA-8.0_Samples/0_Simple/simpleVoteIntrinsics_nvrtc'
find: `/usr/local/cuda-8.0/lib64/stubs': 没有那个文件或目录
>>> WARNING - libcuda.so not found, CUDA Driver is not installed.  Please re-install the driver. <<<
[@] g++ -I../../common/inc -I/usr/local/cuda-8.0/include -o simpleVoteIntrinsics.o -c simpleVoteIntrinsics.cpp
[@] g++ -L/usr/local/cuda-8.0/lib64 -L/usr/local/cuda-8.0/lib64/stubs -o simpleVoteIntrinsics_nvrtc simpleVoteIntrinsics.o -lcuda -lnvrtc
[@] mkdir -p ../../bin/x86_64/linux/release
[@] cp simpleVoteIntrinsics_nvrtc ../../bin/x86_64/linux/release
make[1]: Leaving directory '/home/human-machine/NVIDIA_CUDA-8.0_Samples/0_Simple/simpleVoteIntrinsics_nvrtc'
make[1]: Entering directory '/home/human-machine/NVIDIA_CUDA-8.0_Samples/0_Simple/matrixMul'
"/usr/local/cuda-8.0"/bin/nvcc -ccbin g++ -I../../common/inc  -m64    -gencode arch=compute_20,code=sm_20 -gencode arch=compute_30,code=sm_30 -gencode arch=compute_35,code=sm_35 -gencode arch=compute_37,code=sm_37 -gencode arch=compute_50,code=sm_50 -gencode arch=compute_52,code=sm_52 -gencode arch=compute_60,code=sm_60 -gencode arch=compute_60,code=compute_60 -o matrixMul.o -c matrixMul.cu
nvcc warning : The 'compute_20', 'sm_20', and 'sm_21' architectures are deprecated, and may be removed in a future release (Use -Wno-deprecated-gpu-targets to suppress warning).
cc1plus: fatal error: cuda_runtime.h: 没有那个文件或目录
compilation terminated.
Makefile:250: recipe for target 'matrixMul.o' failed
make[1]: *** [matrixMul.o] Error 1
make[1]: Leaving directory '/home/human-machine/NVIDIA_CUDA-8.0_Samples/0_Simple/matrixMul'
Makefile:52: recipe for target '0_Simple/matrixMul/Makefile.ph_build' failed
make: *** [0_Simple/matrixMul/Makefile.ph_build] Error 2

CUDA9.1+cuDNN7.1 编译测试正常：

https://www.linuxidc.com/Linux/2017-08/146391.htm :参考官方文档，干货
https://blog.csdn.net/weixin_32820767/article/details/80421913
http://www.sohu.com/a/225953058_491081

Ubuntu16.04下安装多版本cuda和cudnn

cuda版本切换

猜你喜欢