Elastic GPU Service:Uninstall an NVIDIA Tesla driver

Uninstall a Tesla driver from Linux

The method for uninstalling a Tesla driver varies based on the installation method of the driver. You must select a method to uninstall your Tesla driver based on the installation scenario of the driver. In this example, a Tesla driver of version 470.161.03 and CUDA 11.4.1 are uninstalled.

Step 1: Uninstall nvidia-fabricmanager

This step is required only when your GPU-accelerated instance belongs to the ebmgn7 or ebmgn7e instance family. In this example, a Tesla driver of version 470.161.03 is uninstalled.

1. Run the following commands to stop nvidia-fabricmanager:

   systemctl stop nvidia-fabricmanager
   systemctl disable nvidia-fabricmanager

2. Run the following commands to uninstall the installation package of nvidia-fabricmanager:

   The commands vary based on your OS.

   • CentOS or Alibaba Cloud Linux

     # Run the following command to view the name of the installation package:
     rpm -qa | grep nvidia-fabric-manager
     
     # Run the following command to uninstall the package:
     rpm -e nvidia-fabric-manager-470.161.03-1.x86_64

   • Ubuntu

     # Run the following command to view the name of the installation package:
     dpkg --list |grep nvidia-fabricmanager
     
     # Run the following command to uninstall the package:
     apt remove nvidia-fabricmanager-470

   Step 2: Uninstall a Tesla driver

   If you use a specific type of installation package such as a .run, .deb, or .rpm installation package to install a Tesla driver on an OS such as Ubuntu, CentOS, Alibaba Cloud Linux, or SUSE, perform the following operations to uninstall the Tesla driver based on the installation scenario of the driver.

   Scenario 1: The Tesla driver is installed by using a .run installation package.

   If the Tesla driver is automatically installed when you create a GPU-accelerated instance, perform the following operations to uninstall the driver by using the .run installation package. In this example, a Tesla diver of version 470.161.03 and CUDA 11.4.1 are uninstalled.

   1. Run the following command to uninstall the Tesla driver:

      /usr/bin/nvidia-uninstall

   2. Run the following commands to uninstall CUDA:

      /usr/local/cuda/bin/cuda-uninstaller
      rm -rf /usr/local/cuda-11.4

      

      Note

      The commands that you run to uninstall a driver may vary based on the CUDA version. If you cannot find the cuda-uninstaller file, check whether a file whose name contains the uninstall_cuda prefix exists in the /usr/local/cuda/bin/ directory. If the file whose name contains the uninstall_cuda prefix exists, replace cuda-uninstaller in the preceding command with the name of the file.

   3. Run the following command to restart the instance:

      reboot

   Scenario 2: The Tesla driver is installed by using a .deb installation package.

   If you use a .deb installation package to install a Tesla driver on Ubuntu, perform the following operations to uninstall the driver. In this example, a Tesla driver of version 470.161.03 and CUDA 11.4.1 are uninstalled.

   1. Run the following commands to uninstall CUDA:

      apt-get --purge remove "*cuda*" "*cublas*" "*cufft*" "*cufile*" "*curand*" \
       "*cusolver*" "*cusparse*" "*gds-tools*" "*npp*" "*nvjpeg*" "nsight*" "*nvvm*"
      
      rm -rf /usr/local/cuda-11.4

   2. Run the following command to uninstall the Tesla driver:

      apt-get --purge remove "*nvidia*" "libxnvctrl*"

   3. Run the following command to uninstall and clean all installation packages:

      apt-get autoremove

   4. Run the following command to restart the instance:

      reboot

   Scenario 3: The Tesla driver is installed by using a .rpm installation package.

   The method to uninstall a Tesla driver varies based on the OS on which you use a .rpm installation package to install the driver.

   CentOS 7 or Alibaba Cloud Linux 2

   If you install a Tesla driver by using a .rpm installation package on CentOS 7 or Alibaba Cloud Linux 2, perform the following operations to uninstall the driver. In this example, a Tesla driver of version 470.161.03 and CUDA 11.4.1 are uninstalled.

   1. Run the following commands to uninstall CUDA:

      yum remove "cuda*" "*cublas*" "*cufft*" "*cufile*" "*curand*" \
       "*cusolver*" "*cusparse*" "*gds-tools*" "*npp*" "*nvjpeg*" "nsight*" "*nvvm*"
      
      rm -rf /usr/local/cuda-11.4

   2. Run the following command to uninstall the Tesla driver:

      yum remove "*nvidia*"

   3. Run the following command to restart the instance to complete the uninstalling process:

      reboot

   CentOS 8 or Alibaba Cloud Linux 3

   If you install a Tesla driver by using a .rpm installation package on CentOS 8 or Alibaba Cloud Linux 3, perform the following operations to uninstall the driver. In this example, a Tesla driver of version 470.161.03 and CUDA 11.4.1 are uninstalled.

   1. Run the following commands to uninstall CUDA:

      dnf remove "cuda*" "*cublas*" "*cufft*" "*cufile*" "*curand*" \
       "*cusolver*" "*cusparse*" "*gds-tools*" "*npp*" "*nvjpeg*" "nsight*" "*nvvm*"
      
      rm -rf /usr/local/cuda-11.4

   2. Run the following command to uninstall the Tesla driver:

      dnf module remove --all nvidia-driver

   3. Run the following command to reset the Tesla driver:

       dnf module reset nvidia-driver

   4. Run the following command to restart the instance:

      reboot

   SUSE

   If you install a Tesla driver by using a .rpm installation package on SUSE, perform the following operations to uninstall the driver. In this example, a Tesla driver of version 470.161.03 and CUDA 11.4.1 are uninstalled.

   1. Run the following commands to uninstall CUDA:

      zypper remove "cuda*" "*cublas*" "*cufft*" "*cufile*" "*curand*" \
       "*cusolver*" "*cusparse*" "*gds-tools*" "*npp*" "*nvjpeg*" "nsight*" "*nvvm*"
      
      rm -rf /usr/local/cuda-11.4

   2. Run the following command to uninstall the Tesla driver:

      zypper remove "*nvidia*"

   3. Run the following command to restart the instance:

      reboot