導入具有GPU支持的TensorFlow時DLL加載失敗

Question

我正在嘗試在Windows 10上安裝具有GPU支持的TensorFlow，但是在導入它時出現錯誤（如下所示）。 CPU版本工作正常。

我有

• 通過pip安裝了tensorflow-gpu
• 使用GeForce Experience更新了我的GTX 1050的NVidia驅動程序
• 通過NVidia的網絡安裝程序安裝了CUDA 10.1
• 已安裝cuDNN 7.5.0.56，負責復制正確CUDA文件夾中的每個文件
• 通過zip方法安裝了TensorRT 5.1.2.2，並再次在CUDA中復制了相關的DLL

這是錯誤：

> python
Python 3.7.2 (tags/v3.7.2:9a3ffc0492, Dec 23 2018, 23:09:28) [MSC v.1916 64 bit (AMD64)] on win32
Type "help", "copyright", "credits" or "license" for more information.
>>> import tensorflow
Traceback (most recent call last):
  File "C:\Users\user\AppData\Local\Programs\Python\Python37\lib\site-packages\tensorflow\python\pywrap_tensorflow.py", line 58, in <module>
    from tensorflow.python.pywrap_tensorflow_internal import *
  File "C:\Users\user\AppData\Local\Programs\Python\Python37\lib\site-packages\tensorflow\python\pywrap_tensorflow_internal.py", line 28, in <module>
    _pywrap_tensorflow_internal = swig_import_helper()
  File "C:\Users\user\AppData\Local\Programs\Python\Python37\lib\site-packages\tensorflow\python\pywrap_tensorflow_internal.py", line 24, in swig_import_helper
    _mod = imp.load_module('_pywrap_tensorflow_internal', fp, pathname, description)
  File "C:\Users\user\AppData\Local\Programs\Python\Python37\lib\imp.py", line 242, in load_module
    return load_dynamic(name, filename, file)
  File "C:\Users\user\AppData\Local\Programs\Python\Python37\lib\imp.py", line 342, in load_dynamic
    return _load(spec)
ImportError: DLL load failed: The specified module could not be found.

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "C:\Users\user\AppData\Local\Programs\Python\Python37\lib\site-packages\tensorflow\__init__.py", line 24, in <module>
    from tensorflow.python import pywrap_tensorflow  # pylint: disable=unused-import
  File "C:\Users\user\AppData\Local\Programs\Python\Python37\lib\site-packages\tensorflow\python\__init__.py", line 49, in <module>
    from tensorflow.python import pywrap_tensorflow
  File "C:\Users\user\AppData\Local\Programs\Python\Python37\lib\site-packages\tensorflow\python\pywrap_tensorflow.py", line 74, in <module>
    raise ImportError(msg)
ImportError: Traceback (most recent call last):
  File "C:\Users\user\AppData\Local\Programs\Python\Python37\lib\site-packages\tensorflow\python\pywrap_tensorflow.py", line 58, in <module>
    from tensorflow.python.pywrap_tensorflow_internal import *
  File "C:\Users\user\AppData\Local\Programs\Python\Python37\lib\site-packages\tensorflow\python\pywrap_tensorflow_internal.py", line 28, in <module>
    _pywrap_tensorflow_internal = swig_import_helper()
  File "C:\Users\user\AppData\Local\Programs\Python\Python37\lib\site-packages\tensorflow\python\pywrap_tensorflow_internal.py", line 24, in swig_import_helper
    _mod = imp.load_module('_pywrap_tensorflow_internal', fp, pathname, description)
  File "C:\Users\user\AppData\Local\Programs\Python\Python37\lib\imp.py", line 242, in load_module
    return load_dynamic(name, filename, file)
  File "C:\Users\user\AppData\Local\Programs\Python\Python37\lib\imp.py", line 342, in load_dynamic
    return _load(spec)
ImportError: DLL load failed: The specified module could not be found.


Failed to load the native TensorFlow runtime.

See https://www.tensorflow.org/install/errors

for some common reasons and solutions.  Include the entire stack trace
above this error message when asking for help.

我的PATH環境變量（我刪除了所有非Microsoft和非NVidia條目）：

> echo $env:PATH.Split(';')
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.1\bin
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.1\libnvvp
C:\Program Files\Microsoft MPI\Bin\
C:\WINDOWS\system32
C:\WINDOWS
C:\WINDOWS\System32\Wbem
C:\WINDOWS\System32\WindowsPowerShell\v1.0\
C:\WINDOWS\System32\OpenSSH\
C:\Program Files\Microsoft SQL Server\130\Tools\Binn\
C:\Program Files\dotnet\
C:\Program Files (x86)\dotnet\
C:\Program Files (x86)\NVIDIA Corporation\PhysX\Common
C:\Program Files\NVIDIA Corporation\Nsight Compute 2019.1\
C:\Program Files\NVIDIA Corporation\NVIDIA NvDLISR
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.1\extras\CUPTI\lib64
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.1\include
C:\Program Files\TensorRT-5.1.2.2\lib
C:\Program Files\TensorRT-5.1.2.2\bin
C:\tools\cuda\bin

TensorRT安裝在C:\\Program Files\\TensorRT-5.1.2.2 。

CUDA bin文件夾的內容：

PS C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.1\bin> ls


    Directory: C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.1\bin


Mode                LastWriteTime         Length Name
----                -------------         ------ ----
d-----       18/04/2019     12:39                crt
-a----       09/02/2019     06:57         202752 bin2c.exe
-a----       09/02/2019     06:57       74850816 cublas64_10.dll
-a----       09/02/2019     06:57       36055552 cublasLt64_10.dll
-a----       09/02/2019     06:57         374272 cuda-memcheck.exe
-a----       09/02/2019     06:57        4320256 cudafe++.exe
-a----       09/02/2019     06:57         339968 cudart32_101.dll
-a----       09/02/2019     06:57         408064 cudart64_101.dll
-a----       18/04/2019     15:19      392566784 cudnn64_7.dll
-a----       09/02/2019     06:57      115644416 cufft64_10.dll
-a----       09/02/2019     06:57         225792 cufftw64_10.dll
-a----       09/02/2019     06:57        4785664 cuinj64_101.dll
-a----       09/02/2019     06:57        2074624 cuobjdump.exe
-a----       09/02/2019     06:57       49030656 curand64_10.dll
-a----       09/02/2019     06:57      171052032 cusolver64_10.dll
-a----       09/02/2019     06:57       87930368 cusparse64_10.dll
-a----       09/02/2019     06:57         285184 fatbinary.exe
-a----       09/02/2019     06:57        1323008 gpu-library-advisor.exe
-a----       09/02/2019     06:57         233472 nppc64_10.dll
-a----       09/02/2019     06:57       10049024 nppial64_10.dll
-a----       09/02/2019     06:57        3719680 nppicc64_10.dll
-a----       09/02/2019     06:57        1003008 nppicom64_10.dll
-a----       09/02/2019     06:57        7227904 nppidei64_10.dll
-a----       09/02/2019     06:57       46016512 nppif64_10.dll
-a----       09/02/2019     06:57       24603136 nppig64_10.dll
-a----       09/02/2019     06:57        5848064 nppim64_10.dll
-a----       09/02/2019     06:57       17426944 nppist64_10.dll
-a----       09/02/2019     06:57         194560 nppisu64_10.dll
-a----       09/02/2019     06:57        2610688 nppitc64_10.dll
-a----       09/02/2019     06:57        8202240 npps64_10.dll
-a----       09/02/2019     06:57         247296 nvblas64_10.dll
-a----       09/02/2019     06:57         384000 nvcc.exe
-a----       09/02/2019     06:57            310 nvcc.profile
-a----       09/02/2019     06:57       22919168 nvdisasm.exe
-a----       09/02/2019     06:57      149298688 nvgraph64_10.dll
-a----       18/04/2019     16:46      131539456 nvinfer.dll
-a----       18/04/2019     16:46        3663360 nvinfer_plugin.dll
-a----       09/02/2019     06:57        5960192 nvlink.exe
-a----       18/04/2019     16:46        4457472 nvonnxparser.dll
-a----       18/04/2019     16:46        2447872 nvparsers.dll
-a----       09/02/2019     06:57        4243456 nvprof.exe
-a----       09/02/2019     06:57         221696 nvprune.exe
-a----       09/02/2019     06:57        4580352 nvrtc-builtins64_101.dll
-a----       09/02/2019     06:57       15659520 nvrtc64_101_0.dll
-a----       09/02/2019     06:57             53 nvvp.bat
-a----       09/02/2019     06:57        5838336 ptxas.exe

和C:\\tools的內容：

PS C:\tools> tree /F
Folder PATH listing
Volume serial number is 6A54-3EA7
C:.
└───cuda
    │   NVIDIA_SLA_cuDNN_Support.txt
    │
    ├───bin
    │       cudnn64_7.dll
    │
    ├───include
    │       cudnn.h
    │
    └───lib
        └───x64
                cudnn.lib

Answer 1

問題是我安裝了CUDA 10.1而不是10.0。 TensorFlow尚未與10.1兼容。

Answer 2

是的，TensorFlow的某些版本不支持CUDA 10.1，在這種情況下，CUDA 10.0或更低版本將解決您的問題。

Answer 3

這就是我解決的方法，如果您使用的是pip或pip3，則第一次成功的機會就更少。

• 卸載Python，CUDA，CuNN，Tensorflow-gpu（如果已安裝）
• 安裝ANACONDA或MINICONDA
• 打開ANACONDA提示符並運行以下命令：

conda創建--name tf_gpu tensorflow-gpu

• 這將創建一個環境tf_gpu，其中將安裝所有兼容版本的Python，CUDA，CuNN和Tensorflow
• 安裝完所有軟件包后，打開ANACONDA提示符並鍵入以下命令

康達激活tf_gpu

• 這會將conda'（base）'更改為'（tf_gpu）'環境
• 然后像往常一樣鍵入python以激活python提示符
• 如果要停用“ tf_gpu”環境

康達停用