快速安装tensforflow-gpu version1.14

骑猪看日落 2023-02-20 02:34 1阅读 0赞

项目需要，再次配置新环境，记录一下，

### **创建环境** ###

conda create --name tf114gpu python=3.7

python最新版是3.8.3，但我一直用3.6，3.7习惯了，这里用3.7是因为要支持后面的3.7版的tensorflow。

如果你已经安装了3.6或其他版本，用  
conda install python=3.7.7  
命令可以把python更换成3.7版。

当然，后面你也可以选择3.6的tensorflow包。

### **安装cudatoolkit+cudnn** ###

直接指定版本号，  
conda install -c anaconda cudatoolkit=10.1

这里要注意，如果要使用anaconda默认的tensorflow安装，则必须 使用10.0版本，且要和你的驱动保持一致。  
因为我已经在本机上安装了10.1的cuda驱动，所以我这里直接使用10.1

后续安装 cudnn（参考：https://anaconda.org/anaconda/cudnn），此时cudnn可以不用指定版本号，  
conda install -c anaconda cudnn  
版本会自动匹配，安装时有版本提示，提示中可以看到相兼容的版本信息。

### **安装tensorflow-gpu=1.14/1.15** ###

**(1) 默认安装**

这里要注意，首先，你使用默认的安装的话，tensorflow1.14只接受cuda10.0，不接受10.1。

conda install -c anaconda tensorflow-gpu=1.14

此时，在列出的安装包中你会发现，tensorflow只支持cuda10.0，不支持cuda10.1等驱动，这是个糟糕的地方。

参考：  
[https://forums.developer.nvidia.com/t/cuda-10-1-tensorflow-1-13/70940/12][https_forums.developer.nvidia.com_t_cuda-10-1-tensorflow-1-13_70940_12]

在tensorflow的官方网站上，也明确指出，CUDA 10.1需要 (TensorFlow >= 2.1.0)  
[https://www.tensorflow.org/install/gpu][https_www.tensorflow.org_install_gpu]

**Software requirements**

The following NVIDIA® software must be installed on your system:

*  [NVIDIA® GPU drivers][NVIDIA_ GPU drivers] —CUDA 10.1 requires 418.x or higher.
 *  [CUDA® Toolkit][CUDA_ Toolkit] —TensorFlow supports CUDA 10.1 (TensorFlow >= 2.1.0)
 *  [CUPTI][] ships with the CUDA Toolkit.
 *  [cuDNN SDK][] (>= 7.6)
 *  *(Optional)* [TensorRT 6.0][] to improve latency and throughput for inference on some models.

**(2) 安装tensorflow1.14+cuda10.1**

我这里没有按anaconda 默认的方式安装，而是在【[github的这个地方][github]】

可以用这个，  
[https://github.com/fo40225/tensorflow-windows-wheel/blob/master/1.14.0/py37/GPU/cuda101cudnn76sse2/tensorflow\_gpu-1.14.0-cp37-cp37m-win\_amd64.whl][https_github.com_fo40225_tensorflow-windows-wheel_blob_master_1.14.0_py37_GPU_cuda101cudnn76sse2_tensorflow_gpu-1.14.0-cp37-cp37m-win_amd64.whl]  
下载了支持cuda10.1的tensorflow版本，然后用命令安装，  
pip install D:\\downloaded\_package\\tensorflow\_gpu-1.14.0-cp37-cp37m-win\_amd64.whl

我安装的是这个，和CPU有关，  
[https://github.com/fo40225/tensorflow-windows-wheel/tree/master/1.14.0/py37/GPU/cuda101cudnn76avx2][https_github.com_fo40225_tensorflow-windows-wheel_tree_master_1.14.0_py37_GPU_cuda101cudnn76avx2]  
下载下来后用7-zip解压，然后安装即可，干脆把安装log也贴出来吧，顺便测试了一下

(tfgpu114) C:\Users\SpaceVision>pip install  D:\mTensorflow\tensorflow_gpu-1.14.0+nv-cp37-cp37m-win_amd64\tensorflow_gpu-1.14.0+nv-cp37-cp37m-win_amd64.whl
    Processing d:\mtensorflow\tensorflow_gpu-1.14.0+nv-cp37-cp37m-win_amd64\tensorflow_gpu-1.14.0+nv-cp37-cp37m-win_amd64.whl
    Requirement already satisfied: keras-applications>=1.0.6 in d:\anaconda3\envs\tfgpu114\lib\site-packages (from tensorflow-gpu==1.14.0+nv) (1.0.8)
    Requirement already satisfied: wrapt>=1.11.1 in d:\anaconda3\envs\tfgpu114\lib\site-packages (from tensorflow-gpu==1.14.0+nv) (1.12.1)
    Requirement already satisfied: tensorboard<1.15.0,>=1.14.0 in d:\anaconda3\envs\tfgpu114\lib\site-packages (from tensorflow-gpu==1.14.0+nv) (1.14.0)
    Requirement already satisfied: astor>=0.6.0 in d:\anaconda3\envs\tfgpu114\lib\site-packages (from tensorflow-gpu==1.14.0+nv) (0.8.1)
    Requirement already satisfied: tensorflow-estimator<1.15.0rc0,>=1.14.0rc0 in d:\anaconda3\envs\tfgpu114\lib\site-packages (from tensorflow-gpu==1.14.0+nv) (1.14.0)
    Requirement already satisfied: numpy<2.0,>=1.14.5 in d:\anaconda3\envs\tfgpu114\lib\site-packages (from tensorflow-gpu==1.14.0+nv) (1.16.4)
    Requirement already satisfied: grpcio>=1.8.6 in d:\anaconda3\envs\tfgpu114\lib\site-packages (from tensorflow-gpu==1.14.0+nv) (1.29.0)
    Requirement already satisfied: termcolor>=1.1.0 in d:\anaconda3\envs\tfgpu114\lib\site-packages (from tensorflow-gpu==1.14.0+nv) (1.1.0)
    Requirement already satisfied: absl-py>=0.7.0 in d:\anaconda3\envs\tfgpu114\lib\site-packages (from tensorflow-gpu==1.14.0+nv) (0.9.0)
    Requirement already satisfied: google-pasta>=0.1.6 in d:\anaconda3\envs\tfgpu114\lib\site-packages (from tensorflow-gpu==1.14.0+nv) (0.2.0)
    Requirement already satisfied: gast>=0.2.0 in d:\anaconda3\envs\tfgpu114\lib\site-packages (from tensorflow-gpu==1.14.0+nv) (0.3.3)
    Requirement already satisfied: keras-preprocessing>=1.0.5 in d:\anaconda3\envs\tfgpu114\lib\site-packages (from tensorflow-gpu==1.14.0+nv) (1.1.2)
    Requirement already satisfied: six>=1.10.0 in d:\anaconda3\envs\tfgpu114\lib\site-packages (from tensorflow-gpu==1.14.0+nv) (1.15.0)
    Requirement already satisfied: protobuf>=3.6.1 in d:\anaconda3\envs\tfgpu114\lib\site-packages (from tensorflow-gpu==1.14.0+nv) (3.12.2)
    Requirement already satisfied: wheel>=0.26 in d:\anaconda3\envs\tfgpu114\lib\site-packages (from tensorflow-gpu==1.14.0+nv) (0.34.2)
    Requirement already satisfied: h5py in d:\anaconda3\envs\tfgpu114\lib\site-packages (from keras-applications>=1.0.6->tensorflow-gpu==1.14.0+nv) (2.10.0)
    Requirement already satisfied: setuptools>=41.0.0 in d:\anaconda3\envs\tfgpu114\lib\site-packages (from tensorboard<1.15.0,>=1.14.0->tensorflow-gpu==1.14.0+nv) (47.3.0.post20200616)
    Requirement already satisfied: markdown>=2.6.8 in d:\anaconda3\envs\tfgpu114\lib\site-packages (from tensorboard<1.15.0,>=1.14.0->tensorflow-gpu==1.14.0+nv) (3.2.2)
    Requirement already satisfied: werkzeug>=0.11.15 in d:\anaconda3\envs\tfgpu114\lib\site-packages (from tensorboard<1.15.0,>=1.14.0->tensorflow-gpu==1.14.0+nv) (1.0.1)
    Requirement already satisfied: importlib-metadata; python_version < "3.8" in d:\anaconda3\envs\tfgpu114\lib\site-packages (from markdown>=2.6.8->tensorboard<1.15.0,>=1.14.0->tensorflow-gpu==1.14.0+nv) (1.6.1)
    Requirement already satisfied: zipp>=0.5 in d:\anaconda3\envs\tfgpu114\lib\site-packages (from importlib-metadata; python_version < "3.8"->markdown>=2.6.8->tensorboard<1.15.0,>=1.14.0->tensorflow-gpu==1.14.0+nv) (3.1.0)
    Installing collected packages: tensorflow-gpu
    Successfully installed tensorflow-gpu-1.14.0+nv
    
    (tfgpu114) C:\Users\SpaceVision>python
    Python 3.7.7 (default, May  6 2020, 11:45:54) [MSC v.1916 64 bit (AMD64)] :: Anaconda, Inc. on win32
    Type "help", "copyright", "credits" or "license" for more information.
    >>> import tensorflow as tf
    >>> print(tf.__version__)
    1.14.0
    >>> sess = tf.compat.v1.Session()
    2020-06-22 12:50:22.879257: I tensorflow/stream_executor/platform/default/dso_loader.cc:42] Successfully opened dynamic library nvcuda.dll
    2020-06-22 12:50:23.127073: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1640] Found device 0 with properties:
    name: GeForce GTX 1080 Ti major: 6 minor: 1 memoryClockRate(GHz): 1.683
    pciBusID: 0000:01:00.0
    2020-06-22 12:50:23.166288: I tensorflow/stream_executor/platform/default/dlopen_checker_stub.cc:25] GPU libraries are statically linked, skip dlopen check.
    2020-06-22 12:50:23.209658: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1717] Ignoring visible gpu device (device: 0, name: GeForce GTX 1080 Ti, pci bus id: 0000:01:00.0, compute capability: 6.1) with Cuda compute capability 6.1. The minimum required Cuda capability is 7.0.
    2020-06-22 12:50:23.290148: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1181] Device interconnect StreamExecutor with strength 1 edge matrix:
    2020-06-22 12:50:23.350693: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1187]
    >>>
    >>> import os
    >>> from tensorflow.python.client import device_lib
    >>> os.environ["TF_CPP_MIN_LOG_LEVEL"] = "99"
    >>> print(device_lib.list_local_devices())
    2020-06-22 12:55:19.209342: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1640] Found device 0 with properties:
    name: GeForce GTX 1080 Ti major: 6 minor: 1 memoryClockRate(GHz): 1.683
    pciBusID: 0000:01:00.0
    2020-06-22 12:55:19.225818: I tensorflow/stream_executor/platform/default/dlopen_checker_stub.cc:25] GPU libraries are statically linked, skip dlopen check.
    2020-06-22 12:55:19.241562: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1717] Ignoring visible gpu device (device: 0, name: GeForce GTX 1080 Ti, pci bus id: 0000:01:00.0, compute capability: 6.1) with Cuda compute capability 6.1. The minimum required Cuda capability is 7.0.
    2020-06-22 12:55:19.388335: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1181] Device interconnect StreamExecutor with strength 1 edge matrix:
    2020-06-22 12:55:19.393806: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1187]      0
    2020-06-22 12:55:19.399103: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1200] 0:   N
    [name: "/device:CPU:0"
    device_type: "CPU"
    memory_limit: 268435456
    locality {
    }
    incarnation: 3946158223506979494
    ]
    >>>

### 安装numpy ###

这里特别拿出来说是因为，tensorflow对numpy的版本要求比较高，不然会报一堆警告出来，参考，  
[https://stackoverflow.com/questions/58662410/which-numpy-versions-are-compatible-with-tensorflow-1-14-0][https_stackoverflow.com_questions_58662410_which-numpy-versions-are-compatible-with-tensorflow-1-14-0]

如果你有多个版本的numpy，你可以检查 一下先，  
`pip show numpy`

`通常新安装的环境numpy版本都比较高。如果你的版本不是1.16.4，那么就要先卸载`  
`pip uninstall numpy`

`然后再安装numpy1.16.4`  
`pip install numpy==1.16.4`

更换版本也可以这样  
pip install -U numpy==1.16.4

[https_forums.developer.nvidia.com_t_cuda-10-1-tensorflow-1-13_70940_12]: https://forums.developer.nvidia.com/t/cuda-10-1-tensorflow-1-13/70940/12
[https_www.tensorflow.org_install_gpu]: https://www.tensorflow.org/install/gpu
[NVIDIA_ GPU drivers]: https://www.nvidia.com/drivers
[CUDA_ Toolkit]: https://developer.nvidia.com/cuda-toolkit-archive
[CUPTI]: http://docs.nvidia.com/cuda/cupti/
[cuDNN SDK]: https://developer.nvidia.com/cudnn
[TensorRT 6.0]: https://docs.nvidia.com/deeplearning/sdk/tensorrt-install-guide/index.html
[github]: https://github.com/fo40225/tensorflow-windows-wheel
[https_github.com_fo40225_tensorflow-windows-wheel_blob_master_1.14.0_py37_GPU_cuda101cudnn76sse2_tensorflow_gpu-1.14.0-cp37-cp37m-win_amd64.whl]: https://github.com/fo40225/tensorflow-windows-wheel/blob/master/1.14.0/py37/GPU/cuda101cudnn76sse2/tensorflow_gpu-1.14.0-cp37-cp37m-win_amd64.whl
[https_github.com_fo40225_tensorflow-windows-wheel_tree_master_1.14.0_py37_GPU_cuda101cudnn76avx2]: https://github.com/fo40225/tensorflow-windows-wheel/tree/master/1.14.0/py37/GPU/cuda101cudnn76avx2
[https_stackoverflow.com_questions_58662410_which-numpy-versions-are-compatible-with-tensorflow-1-14-0]: https://stackoverflow.com/questions/58662410/which-numpy-versions-are-compatible-with-tensorflow-1-14-0