win10使用anaconda将TensorFlow2.1GPU版安装在虚拟环境中
使用anaconda将TensorFlow2.1GPU版安装在虚拟环境中
- 1.打开Anaconda Prompt
- 2.建立一个名为tf2.1（该名称可随便取）的环境，使用Python3.7
- 3.进入tf2.1环境
- 4.安装cuda10.1（因为这里安的TensorFlow是2.1版本，对应的cuda版本是10.1，cudnn版本是7.6，如果要安装别的版本的TensorFlow，需安装对应的cuda和cudnn版本。）
- 5.安装cudnn7.6
- 6.安装TensorFlow-GPU2.1
- 2021.6.30补充
-
- 使用nvcc -V 可以查看系统环境下（本地）的cuda版本。
- 查看conda支持的cuda和cudnn版本
- 测试
因为之前在电脑上安装了PyTorch，装了cuda10.2，网上有各种说法，有说cuda10.2没有对应的TensorFlow版本，所以就用不了TensorFlow了，还有的说可以在本地装两个cuda，一个cuda10.2，一个cuda10.0然后就能使用cuda10.0版本对应的TensorFlow-GPU版了，但我按照方法安装之后，安装是可以安装，但是不能使用GPU加速。
跑测试代码会显示GPU：false
测试代码如下：
import tensorflow as tf
import os
os.environ['TF_CPP_MIN_LOG_LEVEL']='2'
a = tf.constant(1.)
b = tf.constant(2.)
print(a+b)
print('GPU:', tf.test.is_gpu_available())
报如下错误：
D:\Anaconda3\envs\tf2.1.0\python.exe "C:/Users/1/Desktop/Gesture Recognition/EDenseNet/test2.py"
2021-05-15 17:20:13.842076: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library 'cudart64_101.dll'; dlerror: cudart64_101.dll not found
2021-05-15 17:20:13.842233: I tensorflow/stream_executor/cuda/cudart_stub.cc:29] Ignore above cudart dlerror if you do not have a GPU set up on your machine.
2021-05-15 17:20:15.299935: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library nvcuda.dll
2021-05-15 17:20:16.004743: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1555] Found device 0 with properties:
pciBusID: 0000:01:00.0 name: GeForce MX350 computeCapability: 6.1
coreClock: 1.468GHz coreCount: 5 deviceMemorySize: 2.00GiB deviceMemoryBandwidth: 52.21GiB/s
2021-05-15 17:20:16.005776: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library 'cudart64_101.dll'; dlerror: cudart64_101.dll not found
2021-05-15 17:20:16.088248: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cublas64_10.dll
2021-05-15 17:20:16.128756: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cufft64_10.dll
2021-05-15 17:20:16.142859: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library curand64_10.dll
2021-05-15 17:20:16.194405: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cusolver64_10.dll
tf.Tensor(3.0, shape=(), dtype=float32)
2021-05-15 17:20:16.225556: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cusparse64_10.dll
2021-05-15 17:20:16.226581: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library 'cudnn64_7.dll'; dlerror: cudnn64_7.dll not found
2021-05-15 17:20:16.226706: W tensorflow/core/common_runtime/gpu/gpu_device.cc:1592] Cannot dlopen some GPU libraries. Please make sure the missing libraries mentioned above are installed properly if you would like to use GPU. Follow the guide at https://www.tensorflow.org/install/gpu for how to download and setup the required libraries for your platform.
Skipping registering GPU devices...
2021-05-15 17:20:16.227360: I tensorflow/core/platform/cpu_feature_guard.cc:142] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2
2021-05-15 17:20:16.227895: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1096] Device interconnect StreamExecutor with strength 1 edge matrix:
2021-05-15 17:20:16.227996: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1102]
WARNING:tensorflow:From C:/Users/1/Desktop/Gesture Recognition/EDenseNet/test2.py:10: is_gpu_available (from tensorflow.python.framework.test_util) is deprecated and will be removed in a future version.
Instructions for updating:
Use `tf.config.list_physical_devices('GPU')` instead.
2021-05-15 17:20:16.230236: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1555] Found device 0 with properties:
pciBusID: 0000:01:00.0 name: GeForce MX350 computeCapability: 6.1
coreClock: 1.468GHz coreCount: 5 deviceMemorySize: 2.00GiB deviceMemoryBandwidth: 52.21GiB/s
2021-05-15 17:20:16.231241: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library 'cudart64_101.dll'; dlerror: cudart64_101.dll not found
2021-05-15 17:20:16.231361: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cublas64_10.dll
2021-05-15 17:20:16.231450: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cufft64_10.dll
2021-05-15 17:20:16.231538: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library curand64_10.dll
2021-05-15 17:20:16.231627: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cusolver64_10.dll
2021-05-15 17:20:16.231719: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cusparse64_10.dll
2021-05-15 17:20:16.232770: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library 'cudnn64_7.dll'; dlerror: cudnn64_7.dll not found
2021-05-15 17:20:16.232900: W tensorflow/core/common_runtime/gpu/gpu_device.cc:1592] Cannot dlopen some GPU libraries. Please make sure the missing libraries mentioned above are installed properly if you would like to use GPU. Follow the guide at https://www.tensorflow.org/install/gpu for how to download and setup the required libraries for your platform.
Skipping registering GPU devices...
**GPU: False**
2021-05-15 17:20:16.746028: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1096] Device interconnect StreamExecutor with strength 1 edge matrix:
2021-05-15 17:20:16.746122: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1102] 0
2021-05-15 17:20:16.746174: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1115] 0: N
Process finished with exit code 0
由上图看出虽然装的是GPU版，但是程序并不能调用到GPU。
后来发现可以一个方法可以安装TensorFlow-GPU版：之前我一直都是在本地计算机安装cuda和cudnn，后面发现不在本地安装cuda和cudnn，而是在anaconda新建的虚拟环境中安装cuda和cudnn，然后再安装对应的TensorFlow-GPU版本，可以work。现记录下来，给别人一个参考，也方便以后自己查看。
1.打开Anaconda Prompt
2.建立一个名为tf2.1（该名称可随便取）的环境，使用Python3.7
conda create --name tf2.1 python=3.7
3.进入tf2.1环境
activate tf2.1
4.安装cuda10.1（因为这里安的TensorFlow是2.1版本，对应的cuda版本是10.1，cudnn版本是7.6，如果要安装别的版本的TensorFlow，需安装对应的cuda和cudnn版本。）
conda install cudatoolkit=10.1
5.安装cudnn7.6
conda install cudnn=7.6
6.安装TensorFlow-GPU2.1
pip install tensorflow-gpu==2.1
安装完成就可以使用TensorFlow2.1GPU版本啦！
因为我们是将TensorFlow-GPU2.1装在虚拟环境中的，cuda和cudnn也是装在虚拟环境中的，所以理论上我们可以再新建一个虚拟环境，装不同版本的TensorFlow，同理pytorch也一样，当然我认为是这样的，我自己也没试过。有需要的朋友可以试一下。
2021.6.30补充
使用nvcc -V 可以查看系统环境下（本地）的cuda版本。
nvcc -V
查看conda支持的cuda和cudnn版本
打开Anaconda Prompt,输入下列命令查看conda目前支持的cudatoolkit版本。(cudatoolkit即cuda)。
conda search cudatoolkit --info
输入下列命令查看conda目前支持的cudnn版本。
conda search cudnn --info
（注：安装cudnn时，如果conda install cudnn不指定版本，系统会自动安装和当前cuda匹配的cudnn）
测试
用conda list看下上面在虚拟环境下安装的包。
conda list
可以看到虚拟环境里的cuda版本为10.1，而本地的cuda版本是10.2。说明虚拟环境中的cuda和本地的是相互独立的。
参考链接：https://blog.csdn.net/qq_41719643/article/details/109777042