tensorflow/stream_executor/cuda/cuda_dnn.cc:378] Loaded runtime CuDNN library: 7301--2019.5.12

安装的cudnn的版本是7.1.0.3，而要求的cudnn版本是7.3.0.0。

将tensorflow版本从1.5换成1.8，顺利运行程序（升级tensorflow版本来解决）

 

ll 命令查看 连接  /usr/local/cuda/lib64下 把对应的 libcudnn.so.7,3,1 连到 libcudnn.so.7 在连到libcudnn.so

sudo ln -sf libcudnn.so.7.3.1 libcudnn.so.7

sudo ln -sf libcudnn.so.7 libcudnn.so

sudo ldconfig

 

然后升级一下tensorflow-gpu即可。

使用 nvidia-smi 命令

$ nvidia-smi

但是这个命令只能显示一次，如果要实时显示，配合watch命令, 让一秒刷新一次

$ watch -n 1 nvidia-smi

Ubuntu16.04下安装多版本cuda和cudnn

2018年06月28日 16:33:58 tiankong_hut 阅读数：517

 i7-7700k + TITAN X + 16G DDR4 2400 + 256G SSD

ubuntu16.0.4+ 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   

查看目录属性:    ls -l 目录名

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

1. 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.

2. 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.

3. F tensorflow/core/kernels/conv_ops.cc:532] Check failed: stream->parent()->GetConvolveAlgorithms(&algorithms)

4. 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 

1. 下载CUDA并安装： 
   各个版本的CUDA ：https://developer.nvidia.com/cuda-toolkit-archive

2. 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 File

1. tar -zxvf 压缩文件名.tar.gz

2. sudo cp cuda/include/cudnn.h /usr/local/cuda-8.0/include

3. sudo cp cuda/lib64/libcudnn* /usr/local/cuda-8.0/lib64

4. 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

1. export PATH=/usr/local/cuda/bin${PATH:+:${PATH}}

2. export LD_LIBRARY_PATH=/usr/local/cuda/lib64${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}

下面的不行：

1. export PATH="$PATH:/usr/local/cuda/bin"

2. 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)

1. export PATH=/usr/local/cuda-9.1/bin:$PATH

2. export LD_LIBRARY_PATH=/usr/local/cuda-9.1/lib64:$LD_LIBRARY_PATH

改为:

1. export PATH=/usr/local/cuda/bin:$PATH

2. export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH

 

从cuda9.1切换到cuda8.0：

1. sudo rm -rf /usr/local/cuda                    #删除之前创建的软链接

2. 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：

1. $ nvcc --version

2. nvcc: NVIDIA (R) Cuda compiler driver

3. Copyright (c) 2005-2017 NVIDIA Corporation

4. Built on Fri_Sep__1_21:08:03_CDT_2017

5. Cuda compilation tools, release 9.0, V9.0.176

which nvcc ：查看nvcc位置

CUDA8.0+cuDNN5.1 未完全安装：

1. Installing the CUDA Toolkit in /usr/local/cuda-8.0 ...

2. Installing the CUDA Samples in /home/human-machine ...

3. Copying samples to /home/human-machine/NVIDIA_CUDA-8.0_Samples now...

4. Finished copying samples.

5.  

6. ===========

7. = Summary =

8. ===========

9.  

10. Driver: Not Selected

11. Toolkit: Installed in /usr/local/cuda-8.0

12. Samples: Installed in /home/human-machine

13.  

14. Please make sure that

15. - PATH includes /usr/local/cuda-8.0/bin

16. - 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

17.  

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

19.  

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

21.  

22. ***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.

23. To install the driver using this installer, run the following command, replacing with the name of this run file:

24. sudo .run -silent -driver

25.  

26. 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。

1. human-machine@humanmachine-System-Product-Name:~/NVIDIA_CUDA-8.0_Samples$ make

2. make[1]: Entering directory '/home/human-machine/NVIDIA_CUDA-8.0_Samples/0_Simple/simpleVoteIntrinsics_nvrtc'

3. find: `/usr/local/cuda-8.0/lib64/stubs': 没有那个文件或目录

4. >>> WARNING - libcuda.so not found, CUDA Driver is not installed. Please re-install the driver. <<<

5. [@] g++ -I../../common/inc -I/usr/local/cuda-8.0/include -o simpleVoteIntrinsics.o -c simpleVoteIntrinsics.cpp

6. [@] g++ -L/usr/local/cuda-8.0/lib64 -L/usr/local/cuda-8.0/lib64/stubs -o simpleVoteIntrinsics_nvrtc simpleVoteIntrinsics.o -lcuda -lnvrtc

7. [@] mkdir -p ../../bin/x86_64/linux/release

8. [@] cp simpleVoteIntrinsics_nvrtc ../../bin/x86_64/linux/release

9. make[1]: Leaving directory '/home/human-machine/NVIDIA_CUDA-8.0_Samples/0_Simple/simpleVoteIntrinsics_nvrtc'

10. make[1]: Entering directory '/home/human-machine/NVIDIA_CUDA-8.0_Samples/0_Simple/matrixMul'

11. "/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

12. 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).

13. cc1plus: fatal error: cuda_runtime.h: 没有那个文件或目录

14. compilation terminated.

15. Makefile:250: recipe for target 'matrixMul.o' failed

16. make[1]: *** [matrixMul.o] Error 1

17. make[1]: Leaving directory '/home/human-machine/NVIDIA_CUDA-8.0_Samples/0_Simple/matrixMul'

18. Makefile:52: recipe for target '0_Simple/matrixMul/Makefile.ph_build' failed

19. 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