How to Completely Remove | Install | Upgrade Cuda, Cudnn & Pytorch in Windows For All GPU Types 2024

hello everyone welcome back to my channel in this video I will show you how you can completely remove the Cuda and CNN from your PC on Windows platform and after that I will show you how you can install or upgrade the newer version of Cuda and CNN in your PC and in the end we will also install py torch with the support of Cuda and CNN and in the very last we will test all of our installations so so please watch this video Until end in order to make sure that you don't miss any steps because we have to navigate through a lot of folders to make sure that we are deleting or installing the things at the right place so if you are new in my Channel please don't forget to hit the Subscribe button and also press the Bell icon to get the updates of the new uploads now we will move towards my computer screen so guys I have already installed the Cuda and CN my windows so to completely remove it first I have to press the window keys and go into add or remove programs and after that you have to search here for the Nvidia just type here Nvidia it will display all the Nidia drivers you have installed right now in your PC we have to uninstall every each of the driver except the Nvidia control panel so first we are going to uninstall the CNN development so to uninstall this just click on uninstall yes and it will open the user interface and you just have to click on uninstall and you have to click on close so the next thing we are going to install the CNN run time just click on uninstall uninstall yes and again you have to click on uninstall and close and similarly we will now go for the samples click on uninstall uninstall yes and uninstall next thing is we are going to install the Cuda tool kit and just click on uninstall uninstall yes click on uninstall it will take few seconds here please have some patience yes the Cuda toolkit has been successfully removed just close it and next we are going to uninstall the frame view SDK just click on uninstall uninstall yes uninstall close and the next thing is GeForce experience click on uninstall uninstalled yes uninstall now again it will take few minutes to completely remove the GeForce experience now gForce has been successfully removed just close it and now we will uninstall the graphics driver just click on uninstall install uninstall yes uninstall now we are back again after the restart next you again have to go into the add or remove programs just open it just type here viia again and uh we have to remove all these components one by one please don't remove the control panel uninstall yes next is inside systems uninstall uninstall yes next is the inside Visual Studio Edition yes uninstall uninstall yes and the last one is physix system uninstall uninstall yes uninstall now this one has been successfully removed just close it and next we have to navigate to over C directory just open the folder here open the C directory go into the program files 86 and look for the Nvidia and Nidia just click on continue and it will uninstall the Nvidia from the program files next we have to go into the program files and we have to uninstall Nvidia Nvidia Corporation Nvidia toolkit just select these three and shift lead yes yes continue yes and next we have to go into the C program data and in program data you again have to remove the Nvidia and Nidia Corporation just select these shift delad yes continue he's saying that the file is open somewhere so to do this you have to open your toask manager to close the file you have to type here Nidia and it is open here and with display container we have to end toask this and now you can click on try again now it is successfully removed so the next we have to go into the AP data you can simply go into the app data you have to go into the roaming and here you have to look for Nidia if it's not there it's okay if it's there you have to again remove the Nvidia folder from here next you have to go into the system environment variables for this you can simply type here environment variables you have to go into the system environment variables environment variables and check that if the path for the Cuda is existed here or not for my for my case we have successfully deleted so the path is not here anymore it mean we are going in the right direction just click on okay okay next we have to upgrade or install over Cuda and CNN and next we are going to install the latest version of Cuda and CNN in our windows for this open your Chrome browser and just type here install Cuda and CN and windows and go into the install CNN windows and here we have to follow all these steps to install the toolkit and the CNN for Windows and the first step is installing Nidia graphic drivers and I'm going to skip this step for now and I'm directly jumping to The Next Step which is installing the Huda toolkit for Windows the reason for this is that because when we are going to install the Cuda toolkit it will automatically install the right version of the Nidia graphic drivers so at this stage we have to make sure that we are installing the right version of the Huda toolkit based on our GPU architecture to check this you can simply go into the CNN sport Matrix open it in a new tab and as they have listed here all the Cuda and CNN versions here and I think the latest version for the toolkit is 12.3 and for the CNN is 9 and next we have to make sure that these versions are compatible with our GPU for this you can have to check the Huda compute capability list and sported Nidia Hardware list and we have to make sure that our device is listed in this list so if you are unsure about the name of your device you can simply press the window key and go into the device manager and go into the display adapters and all of your GPU devices are listed here mine is Nidia GeForce gtx1 1650 TI and next we have to check the compute capability number for our device and the hardware name for this you can open a new tab and just type CA compute capability and go into the Wikipedia and just go down down down and here in this table they have listed the compute capability number and also the architecture name for all the GPU devices next you have to locate your your GPU device in this list mine is gForce gtx1 1650 TI and the capability number for my device is 7.5 and the architecture name is during if your device is not listed in this table don't worry you can still open a new tab and just type Cuda compute and your device name gtx1650 TI just click on it and it will display here the capability number for my device is 7.5 and to check the architecture name you can simply write your name and the name of your device and just press enter and the name for my GPU is touring and as we can see in the sport Matrix my GPU Hardware is listed here and also the compute capability number so it means I am able to install this toolkit version and this CNN version and first we are going to install the toolkit just click on it and open it in a new tab and the latest version is 12.4.2 and to install this just click on it and next you have to select the operating system and your architecture and we are using Windows 11 and we have to install it locally and just click on download it will automatically start the download and you have to wait until the download is complete once your download is completed you have to go into your downloads folder and the file we have downloaded is here Cuda 12.40 this 12.4.2 is the Cuda toolkit version and 55161 is the version of the Nidia graphics driver next we are going to open this just click on yes and click on okay and this process will take a few seconds and after the completion of this process a new Supply screen will automatically open and this flash screen is saying we are going to install the Cuda toolkit version 12.4 and now it is checking the system compatibility it will again take few seconds next you have to agree with the license and continue and we will look what we are going to install for this we are going into custom and click on next as you can see it will install the Cuda drivers and all these drivers are listed here it will install the inside develop one time and visual studio integration with this Visual Studio integration you can use your Cuda and kouan drivers in visual studio and next it is also installing the GeForce experience here and in other components it is installing the physics and in the driver components it is installing the display driver and it is installing the version 55161 we are okay with all these things just click on next default location is okay just click on next and now it is saying that in our PC we don't have the visual studio and we are installing the visual studio integration so some components of the Cuda toolkit will not work theye install Visual Studio to get the full functionality if you face any errors if your Cuda installation failed or something happens during the Cuda installation some of the components will not get installed so please make sure to install the visual studio code and next I'm going to show you how you can install it just click on visual studio and you have to go down and you have to install the community version of the visual studio code 2022 it will automatically start the download next you have to open this file and you have to click on yes and you have to click on continue it will take few seconds to download and installing the installer next here you have to select the right components for your visual studio so I'm going to select the desktop development with C++ because it has msvc tools which is required to build the Cuda and CNN drivers so just click on yes and this tool is very important please please don't uncheck this msvc in order to properly install the Cuda and CNN drivers if you face any problems or if your Cuda driver failed to install you have to recheck your Visual Studio installation and you check you must have this msvc installer and next we are going to click on install and now it will take few minutes to properly install this visual studio and I'm going to uncheck this box because I don't want to start off the installation now over installing is almost done and after the successful installation you will see this kind of prompt now I'm going to close this window and we will go back to our Nidia Cuda graphic interface and now I have already installed this now I will click I understand and wish to continue and I click on next and uh I just click on next and now it is just preparing for the installation it will take few seconds and now it is installing our drivers first it is installing the inside visual Studio addition it will take few minutes here you have to do wait now it is installing the graphics driver now it is installing the GeForce experi and as you can see it is said that install andite for visual studio 222 because we have 2022 V studio in our PC and we don't have 2019 and 2017 that's why it's it is not installing the nide for these Visual Studio but this is okay it means our installation is successful we just click on next and we don't want to launch it and we will just close this window that's it our installation of the Cuda tool kit is done next we are going to install the CNN so to install the CNN for Windows first we have to download the CNN from this website and there are two types of installation graphical and terrible in graphical installation it's a executable file it's very simple we just have to follow the prompt and in the zip file you have to download the zip file and next you have to extract the zip file and after the extract you have to copy files from these folder into these folders and after that you have to go to your environment variables and you have to copy this path in your environment variables and don't forget to change this x with the version of your CNN and that's it so we are going to follow the graphical installation so we have to open this link we have to download the CNN Library just click on it and now it is asking for the operating system Windows 64 bit and we are going to do with the version 10 we have Windows 11 but version 10 will also to work on Windows 11 otherwise you can do the turble installation just click on 10 and we are going to local installation just download this file so I have already downloaded this file so I'm going to cross it I am going into my downloads folder and the file name is CNN 9 just double click on it and click on yes and click on okay and again this process will take few seconds and next we have to agree with the license continue and we are going to express installation but in custom installation you can see the files what we are going to install we are going to install the CNN for CA 12.3 11.8 and CNN samples actually we don't need the C DNN for 11.8 because we have Cuda 12.4 I think this CNN will also work for uh 12.4 just uncheck this and select these two just click on next and we have to use the same paths and just click on next and now after the installation it is saying that we have not selected the CNN run time time for 11.3 it's okay we have installed the cian and run time for L 12.4 and we have also installed the samples that's it just close it and next to verify our Cuda and CNN installation we have to open our Command Prompt and run it as an administrator and just type here nvcc minus minus version it will display the version of your package of your Nvidia Tuda toolkit and it's here it's 12.4 to check the other thing your CNN and Cuda version you can type also NV D SMI and press enter and it is saying that our driver version is 55161 and our Huda version is 12.4 it mean over installation is successful next we are going to install the py with the support of Cuda and CNN to install the py with Coda and codan and Sport we have to open our Chrome and just type here install P torch and press enter go into to start locally and choose the stable version and operating system is Windows package we are going to use pip you can also install it with the K but I'm going to using the piip and the language is python make sure that you have already installed the python in your PC otherwise you have to install the python first and Cuda platform instead of 11.8 we have Cuda 12 actually we have Cuda 12.4 but I think 12.1 also works with the 12.4 we just have to then copy this command and we have to open your command prompt and run it as an administrator yes and just paste it here and press enter now it is installing the Pythor and it will take few minutes around 10 minutes to complete this setup based on our internet speed we have to wait here now the downloading of P torch is almost going to complete next it is installing torch vision and now torch audio now it has downloaded all the prerequisites and the python torch torch vision and torch audio and these are the pre prerequisite requires to build the torch torch vision and torch audio now it is going to install these packages it will take some time here also now after the successful installation you will see this message successfully installed now to verify over installation of P that it runs with Cuda or CNN you can simply open the python you have to import torch and here you can just type you just have to type the right command is torch the or torch Cuda a v a i l a b e and press enter and it is saying through and next we have to see the number of devices for this you can again type torch. ca. device and count and press enter and it is saying one device it means our pie T is successfully built with the support of Cuda and CNN this is it for today I hope you have enjoyed this video in the next video we will see how we can the pie thought with the C++ or the Java and see you next time bye-bye

Channel: Robotics Workshop

Views: 2,047

Keywords: how to remove cuda from windows, how to remove cudnn from windows, remove cuda cudnn in windows, remove cuda windows 11, how to install cuda and cudnn in windows, install cudnn in windows, install pytorch, how to install pytorch in windows, pytorch with cuda and cudnn, Cuda toolkit installation failed, cuda installation failed, visual studio cuda integration, cuda, cudnn, pytorch installation in windows, pytorch windows, unistall cuda and cudnn, delete cuda and cudnn, remove cuda

Id: krAUwYslS8E

Length: 17min 34sec (1054 seconds)

Published: Sat Mar 30 2024