When everyone has a little dream to engage in artificial intelligence , Even when it comes to deep learning , Buy a lot of money to buy one N After the card machine , But found your torch.cuda.is_available() was False！ Burst into tears o(╥﹏╥)o, When you find nothing after working for most of the day , Let you suddenly have the idea of smashing the computer . Yes , This article blog Just to help you solve this puzzle , Let you regain the confidence and motivation of deep learning ！！！

0 Take a look at your GPU Support does not support cuda, Generally speaking N Cards are supported .

0.1 Right click the taskbar , Open Task Manager , Look at the model of your card .

0.2 open N card Official website , Slide to the bottom of the home page of the official website , Find out if your model supports cuda. Some too old machines don't support it. Let's get a new machine. Hee hee ~.

1 View your machine support cuda edition .

1.1 The right mouse button of the current machine will have this N Card control panel , To open it .

1.2 find “ help - system information - Components ”, View supported cuda edition , My is 11.4, Take a piece of paper and write down this version , This is what your machine can support cuda Highest version , Let's install it according to this .

1.3 win+R, Input cmd after , Input nvidia-smi, You can see the following , The first number means the current version of the driver , The second one means cuda Version of , It should be consistent with the previous version .

1.4 Input nvcc -V. You will find that there may not be this command , Because you may not have installed cuda, Don't worry, let's go 3 One step to solve this problem . If the command is useful , It will show you again cuda Version of , Here's the picture . This version must be the same as the one just recorded ！

2 Uninstall the previously installed cuda（ It's not installed cuda Your new machine ignores this ）.

2.1 Compared with some friends who may have installed it once cuda, But the discovery is still useless , That's because the wrong version is installed , We need to uninstall the current cuda. Right click the lower left corner of the desktop win Icon , Open search “ Control panel ”,“ View by ” Switch to “ Small icons ”, And then click “ Program and function ”, Unload the original cuda, The procedures to be uninstalled are as follows , Don't uninstall the rest .

2.2 Then use Fujitsu housekeeper or other software to “ rubbish ”, Clear the registry entry .

3 install cuda

3.1 Go to Official website Download the latest cuda Installation package , Options are as follows , Finally, click download .

3.2 The installation steps are as follows

After system check , Choose the first “ Install the video card driver （ It's graphics driven ） and geforce experience（ Graphics card driver automatic update program ）”, Then continue to , Choose Custom ：

The first one here must tick , The last three are also hooked .

Then there is path selection , You can choose the location you want to install .

3.3 After installation , Right click My computer –> attribute –> Advanced system setup –> environment variable , Has been added to the system variable cuda Of 4 A path , Here's the picture .

3.4 Try the second 1 All steps of step , See if all version information is consistent .

4 install pytorch

4.1 open pytorch Official website , Then configure the options , As shown in the figure , Copy the red box command .

4.2 open Anaconda Prompt, Then input “conda config --remove-key channels”, Restore default source , Tsinghua source has stopped maintenance . Then enter the command pasted above . Waiting for download to complete , Unable to download or too slow , Can only go to This page of the official website Find yourself python edition 、cuda The download package corresponding to the version .

5 Be accomplished ！

Test it , After several weeks, I finally got it done , I believe everyone is smarter than me , It should be possible to do well in half an hour at most (^▽)！

5.1 open Anaconda Promp test ：

5.2 open notebook test ：