Solve the problem that CUDA cannot accelerate GPU in pytoch

solve Pytorch in Cuda unable GPU Speed up the problem

---

---

Preface

In the installation pytorch There will be unusable GPU The problem of acceleration , Just install GPU Version of Pytorch Record the process .
Refer to the post ：https://blog.csdn.net/qq_46126258/article/details/112708781

One 、 Test computer CUDA edition

On computer cmd Enter the following command to view the graphics card driver and CUDA Version information

nvcc -V
nvidia-smi

Let me show you the information ：

NVIDIA-SMI 465.89 Driver Version: 465.89 CUDA Version: 11.3

Judge your own CUDA Whether the version meets the requirements （ Installed CUDA and cuDNN bring into correspondence with , The version number can be compared with 11.3 low ）

---

Two 、 Check the video card cudnn edition

Install well cudnn after , open cudnn Folder , My is on the following path ：
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.0\include
open cudnn.h file , I use notepad++ open
ctrl+F retrieval CUDNN_MAJOR , give the result as follows
cudnn:8.2.1

Check the version corresponding problem ：
cuDNN Official website ：https://developer.nvidia.com/rdp/cudnn-archive

---

3、 ... and 、 Installation and downloading PyTorch

1、 Command line installation

download PyTorch You can check the information on the official website ：
https://pytorch.org/get-started/locally/#no-cuda-1
torch If torch-cpu, Indicates that... Is installed cpu Version of torch.
You can install directly from the installation command given on the official website

pip3 install torch==1.10.2+cu113 torchvision==0.11.3+cu113 torchaudio===0.10.2+cu113 -f https://download.pytorch.org/whl/cu113/torch_stable.html
pip3 install torch==1.10.2+cu113 torchvision==0.11.3+cu113 torchaudio===0.10.2+cu113

You can see that the installed packages are the following three packages

torch == 1.10.2+cu113
torchvision == 0.11.3+cu113
torchaudio === 0.10.2+cu113

But one problem with this is , Installation speed is very slow , So we can adopt the method of offline installation .

2、 Offline installation

Get into Pytorch Installation package website , Download the corresponding package file ：
PyTorch Download Web page ：https://download.pytorch.org/whl/torch_stable.html
The following figure is the same as the above package , In fact, you only need to download torch, The other two bags are smaller .

among ,cu113 Express cuda edition 11.3,torch1.10.2 by torch edition ,cp38 Indicates an adaptation python 3.8.

adopt cmd Download the path of their three packages before entering ;
Install one by one in the virtual environment ：

pip install torch-1.10.2+cu113-cp38-cp38-win_amd64.whl
pip install torchvision-0.11.3+cu113-cp38-cp38-win_amd64.whl
pip install torchaudio-0.10.2+cu113-cp38-cp38-win_amd64.whl

---

Four 、 Running tests

After activating the environment cmd Enter the following statement in ：

python
import torch
torch.cuda.is_available()

Show True That is success