Win10系统中GPU深度学习环境配置记录

运行环境

系统：Win10   

处理器 Intel(R) Core(TM) i7-9700K CPU @ 3.60GHz   3.60 GHz
机带 RAM 16.0 GB
设备 ID A18D4ED3-8CA1-4DC6-A6EF-04A33043A5EF
产品 ID 00342-35285-64508-AAOEM
系统类型 64 位操作系统, 基于 x64 的处理器

显卡：NVIDIA GeForce RTX 2070

驱动程序版本: 30.0.15.1252
驱动程序日期: 2022/4/15
DirectX 版本: 12 (FL 12.1)
物理位置： PCI 总线 1、设备 0、功能 0

专用 GPU 内存 0.6/8.0 GB
共享 GPU 内存 0.0/8.0 GB
GPU 内存 0.6/16.0 GB

配置环境

使用GPU进行深度学习需要安装Cuda CuDNN 以及tensorflow或者pytorch等python深度学习框架。

我们可以通过tensorflow官网找到适配的cuDNN和CUDA的版本，网址为：

https://tensorflow.google.cn/instal/source_windows  现在打开显示报错，原文（地址cuDNN和CUDA的安装_cuda和cudnn安装-CSDN博客）显示对应版本如下图：

这里使用版本：Cuda 11.2   CuDnn 8.1  tensorflow_gpu-2.5.0

Cuda下载与安装

下载地址：人工智能计算领域的领导者 | NVIDIA

CUDA Toolkit 12.2 Update 2 Downloads | NVIDIA Developer ，点击

• 选择合适版本Archive of Previous CUDA Releases

选择对应版本进行下载：

下载就得到Cuda11.2安装包。

下面开始安装，点击安装包安装cuda，文件先解压，然后开始安装。

Cudnn下载与安装

下载地址：cuDNN Archive | NVIDIA Developer

这里注意需要注册账号登录之后才能下载。 

下载完整之后的安装包：

Cudnn安装

首先解压下载后的文件，打开文件夹

将bin中所有文件复制到CUDA的bin文件夹（CUDA默认安装到了C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA）

Tensotflow2 下载

下载地址：tensorflow-gpu · PyPI

由于官网下载速度太慢了，这里在CSDN资源里面下载的，地址：https://download.csdn.net/download/zizhuangzhuang/19339448

 版本名称：tensorflow_gpu-2.5.0-cp37-cp37m-win_amd64.whl，这里手动安装到pycharm环境中。手动安装可以参考以前写的博客：python中如何导入gdal包？_python导入gdal_空中旋转篮球的博客-CSDN博客

安装好之后可以在pycharm中查看到结果。 

环境测试

运行深度学习代码，显示cuda加载成功，但是显示错误“Could not load dynamic library 'cudnn64_8.dll'; dlerror: cudnn64_8.dll not found”

2023-10-03 11:11:58.542341: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library cudart64_110.dll
10249
2023-10-03 11:12:04.352079: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library nvcuda.dll
2023-10-03 11:12:04.430194: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1733] Found device 0 with properties: 
pciBusID: 0000:01:00.0 name: NVIDIA GeForce RTX 2070 computeCapability: 7.5
coreClock: 1.62GHz coreCount: 36 deviceMemorySize: 8.00GiB deviceMemoryBandwidth: 417.29GiB/s
2023-10-03 11:12:04.430800: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library cudart64_110.dll
2023-10-03 11:12:04.577966: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library cublas64_11.dll
2023-10-03 11:12:04.578162: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library cublasLt64_11.dll
2023-10-03 11:12:04.668754: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library cufft64_10.dll
2023-10-03 11:12:04.685764: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library curand64_10.dll
2023-10-03 11:12:04.734948: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library cusolver64_11.dll
2023-10-03 11:12:04.776406: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library cusparse64_11.dll
2023-10-03 11:12:04.781208: W tensorflow/stream_executor/platform/default/dso_loader.cc:64] Could not load dynamic library 'cudnn64_8.dll'; dlerror: cudnn64_8.dll not found
2023-10-03 11:12:04.781436: W tensorflow/core/common_runtime/gpu/gpu_device.cc:1766] Cannot dlopen some GPU libraries. Please make sure the missing libraries mentioned above are

重启一下电脑，使用以下代码监测GPU并指定使用GPU重新进行训练 ：

# 指定使用 GPU 设备
print(tf.__version__)
print('is_gpu_available',tf.test.is_gpu_available)
physical_devices = tf.config.list_physical_devices('GPU')
print(physical_devices)
tf.config.experimental.set_memory_growth(physical_devices[0], True)

运行情况如下，没有报以上错误了：

2.5.0
is_gpu_available 0x000001953DC52C18>
2023-10-03 23:25:54.193660: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library nvcuda.dll
2023-10-03 23:25:54.258133: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1733] Found device 0 with properties: 
pciBusID: 0000:01:00.0 name: NVIDIA GeForce RTX 2070 computeCapability: 7.5
coreClock: 1.62GHz coreCount: 36 deviceMemorySize: 8.00GiB deviceMemoryBandwidth: 417.29GiB/s
2023-10-03 23:25:54.258495: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library cudart64_110.dll
2023-10-03 23:25:54.317482: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library cublas64_11.dll
2023-10-03 23:25:54.317677: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library cublasLt64_11.dll
2023-10-03 23:25:54.344522: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library cufft64_10.dll
2023-10-03 23:25:54.352334: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library curand64_10.dll
2023-10-03 23:25:54.375640: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library cusolver64_11.dll
[PhysicalDevice(name='/physical_device:GPU:0', device_type='GPU')]
2023-10-03 23:25:54.397368: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library cusparse64_11.dll
2023-10-03 23:25:54.400881: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library cudnn64_8.dll
2023-10-03 23:25:54.401098: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1871] Adding visible gpu devices: 0
Found 10249 files belonging to 16 classes.
Using 3075 files for training.
2023-10-03 23:25:54.764308: I tensorflow/core/platform/cpu_feature_guard.cc:142] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations:  AVX AVX2
To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.
2023-10-03 23:25:54.765206: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1733] Found device 0 with properties: 
pciBusID: 0000:01:00.0 name: NVIDIA GeForce RTX 2070 computeCapability: 7.5
coreClock: 1.62GHz coreCount: 36 deviceMemorySize: 8.00GiB deviceMemoryBandwidth: 417.29GiB/s
2023-10-03 23:25:54.765556: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1871] Adding visible gpu devices: 0
2023-10-03 23:25:55.259383: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1258] Device interconnect StreamExecutor with strength 1 edge matrix:
2023-10-03 23:25:55.259576: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1264]      0 
2023-10-03 23:25:55.259684: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1277] 0:   N 
2023-10-03 23:25:55.260261: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1418] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 6001 MB memory) -> physical GPU (device: 0, name: NVIDIA GeForce RTX 2070, pci bus id: 0000:01:00.0, compute capability: 7.5)

训练速度有很大提升：训练参数：Trainable params: 11,150,544

CPU状态及训练速度：

Epoch 1/50
257/257 [==============================] - 524s 2sep - loss: 1.4153 - accuracy: 0.5106 - val_loss: 2.0161 - val_accuracy: 0.4628

GPU状态及速度，GPU温度急剧上升，很快就升到80摄氏度以上。GPU训练速度提升了进10倍。

257/257 [==============================] - 66s 209ms/step - loss: 1.3817 - accuracy: 0.5200 - val_loss: 1.6644 - val_accuracy: 0.5265
Epoch 2/50
257/257 [==============================] - 52s 201ms/step - loss: 0.9899 - accuracy: 0.6371 - val_loss: 3.9179 - val_accuracy: 0.2101
Epoch 3/50