Skip to content

Latest commit

 

History

History
89 lines (78 loc) · 5.95 KB

README.md

File metadata and controls

89 lines (78 loc) · 5.95 KB
Raw

Ubuntu-Set-up-Nvidia-cuda-conda-pytorch

a short guide to solving black screen after installing Nvidia-driver on Ubuntu

Step 1. install ubuntu

  1. download ubuntu image to your flash device
  2. download rufus to create the bootable USB

Warning (10/27/2023): The current ubuntu22.04LTS image is not stable, it can't correctly install. But ubuntu20.04LTS and ubuntu23.04 is stable. You can update to ubuntu22.04 once you install ubuntu20.04, but it will cause nvidia-cuda-toolkit needs update. Use chatGPT to help you.

Black screen

Sometimes, you get a black screen and it's very annoying. To deal with the black screen:

  1. press ctrl + shift + ESC + F1 or something else when booting and try to get into the GNU-GRUB window. Note it's not the BIOS and you need to try several times to succeed.
  2. select the advanced option for ubuntu
  3. select the second term (recovery mode) and you are now in the recovery menu.
  4. enter the root
  5. change the grub setting with sudo nano \etc\default\grub and add 'nomodeset' before the quite splash and use Ctrl + x to save. The nomodeset will use the integrated gpu of the motherboard to boot your computer.
  6. enter exit to exit from the root and you are now in the recovery menu.
  7. enter the grub and it will update grub file that you modified in step 5
  8. enter the resume to boot the computer

Or, if you find \etc\default\grub is an empty file and you can not find any solution, the blackscreen may be caused by the wrong version of nvidia-driver. Be cautious when you install third-party version of nvidia-driver, this may cause the blackscreen problem. In your GNU-GRUB window, you should be able to log in root account and download the right version of nvidia-driver, which can fix the blackscreen problem.

you should be good to boot your computer now.

Step 2. install the nvidia-driver

  1. sudo apt update + sudo apt upgrade + sudo apt install gcc + sudo apt install build-essential
  2. sudo apt install nvidia-driver-535 nvidia-dkms-535 Change the version number [535] of the driver that fits your gpu.
  3. ((don't need for now) )remove the nomodeset from the grub with sudo nano \etc\default\grub and update your grub with sudo update-grub. Note that without step 3, you cannot reboot your computer.
  4. reboot your computer and you can check whether the driver is installed with nvidia-smi

Step 3. install cuda. check this video before installing. You can install any version instead of the one shown in the video.

  1. download the cuda toolkit that fits your GPU driver and OS. ( or wget https://developer.download.nvidia.com/compute/cuda/11.8.0/local_installers/cuda_11.8.0_520.61.05_linux.run)

  2. select runfile (local) as the Installer Type and it will generate the Installation Instructions. Note that it will be better to have chmod +x as suggested in the video. Just select install the cuda without the driver, the sample, and the document, as shown in the video. (or sudo sh cuda_11.8.0_520.61.05_linux.run --toolkit --silent --override)

  3. sudo nano ~/.bashrc and add the path to your bashrc file. Note change the 11.1 for your installed cuda version

    export PATH=/usr/local/cuda-11.8/bin${PATH:+:${PATH}}
export LD_LIBRARY_PATH=/usr/local/cuda-11.8/lib64${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}

    The below are out of date fro cuda 11.1

     export PATH=$PATH:/usr/local/cuda-11.1/bin 
 export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda-11.1/lib64
 export CUDADIR=/usr/local/cuda-11.1

  4. source ~/.bashrc and you should be good to get the cuda installed. You can check it with nvcc -V with a new terminal.

Step 4. install cudann (to modify)

  1. download the cudnn. It needs you to login in and verify your email.
  2. download the Deb file that fit your cuda and OS [not recommended!]. Or install it with the compressed file as show in the video above. 
        sudo dpkg -i cudnn-local-repo-ubuntu2004-8.6.0.163_1.0-1_amd64.deb
    sudo cp /var/cudnn-local-repo-*/cudnn-local-*-keyring.gpg /usr/share/keyrings/
  3. install the cuDNN runtime library, the developer library, and the code samples library. 
    sudo apt update
sudo apt upgrade
sudo apt-get install libcudnn8=8.6.0.163-1+cuda11.8
sudo apt-get install libcudnn8-dev=8.6.0.163-1+cuda11.8
sudo apt-get install libcudnn8-samples=8.6.0.163-1+cuda11.8
  4. you can check whether cudnn is installed with sudo apt search cudnn | grep installed if you installed it with .deb. You can also check it with the stackflow answers.
function lib_installed() { /sbin/ldconfig -N -v $(sed 's/:/ /' <<< $LD_LIBRARY_PATH) 2>/dev/null | grep $1; }
function check() { lib_installed $1 && echo "$1 is installed" || echo "ERROR: $1 is NOT installed"; }
check libcudnn

Step 5. install the conda

  1. download conda from the website and install it.
  2. if you'd prefer that conda's base environment not be activated on startup, you can use conda config --set auto_activate_base false.
  3. create and activate the conda environmnet with a specific python version. 
    conda create -n first_env python=3.7
conda activate first_env

Step 6. install pytorch

  1. make sure your conda env is activated recommended
  2. go to pytorch website and search the pytorch version that fits your cuda
  3. install the pytorch with pip
  4. you can check whether it got installed with 
    python -c "import torch; print(torch.__version__)"
python -c "import torch; print(torch.version.cuda)"