Changes

• Updated driver versions:
  • Prod:
    • Standard: 580.126.20 -> 595.71.05
    • RTX: 580.126.09-grid -> 595.71.05-grid
    • CUDA Version: 13.0.2 -> 13.2.1
  • LTS:
    • Standard: 580.126.20 -> 580.159.03
• vGPU support (fractional VMs with virtual GPUs).
• Added list_driver_version command to display a list of available driver versions.
• Added --force-version flag that allows to specify the version of driver to be installed.
• Added --only-dependencies flag that stops the driver installation process after all dependencies are installed. It allows to prepare a system for swift driver installation later.

Requirements

The system on which you want to run the script needs to meet the following
requirements:

• Python interpreter in version 3.8 or newer installed.
• Access to Google Cloud Storage.
• Access to distribution specific repositories, to perform required updates and dependency installations.

Running the tool

The cuda_installer.pyz script needs to be executed with root privileges to install drivers or CUDA Toolkit
(for example sudo python3 cuda_installer.pyz install_cuda).

Note: During the installation the script will trigger system reboots. After a
reboot, the script needs to be started again to continue the installation process.

After successfully installation, the tool will restart your system once more to make
sure everything is initialized properly and working system-wide.

You can use the following script as a startup script for your instance, to make the installation automatic:

#!/bin/bash
if test -f /opt/google/cuda-installer
then
  exit
fi

mkdir -p /opt/google/cuda-installer/
cd /opt/google/cuda-installer/ || exit

curl -fSsL -O https://storage.googleapis.com/compute-gpu-installation-us/installer/v1.9.0/cuda_installer.pyz
python3 cuda_installer.pyz install_cuda

Changes

• Updated driver versions:
  • Prod:
    • Standard: 580.105.08 -> 580.126.20
    • RTX: 580.105.08-grid -> 580.126.09-grid
    • CUDA Version: 13.0.1 -> 13.0.2
  • NFB:
    • CUDA Version: 13.1.0 -> 13.1.1
  • LTS:
    • Standard: 580.105.08 -> 580.126.20
• Fixed installation problems on Debian and Ubuntu in repository mode.

Requirements

The system on which you want to run the script needs to meet the following
requirements:

• Python interpreter in version 3.8 or newer installed.
• Access to Google Cloud Storage.
• Access to distribution specific repositories, to perform required updates and dependency installations.

Running the tool

The cuda_installer.pyz script needs to be executed with root privileges to install drivers or CUDA Toolkit
(for example sudo python3 cuda_installer.pyz install_cuda).

Note: During the installation the script will trigger system reboots. After a
reboot, the script needs to be started again to continue the installation process.

After successfully installation, the tool will restart your system once more to make
sure everything is initialized properly and working system-wide.

You can use the following script as a startup script for your instance, to make the installation automatic:

#!/bin/bash
if test -f /opt/google/cuda-installer
then
  exit
fi

mkdir -p /opt/google/cuda-installer/
cd /opt/google/cuda-installer/ || exit

curl -fSsL -O https://storage.googleapis.com/compute-gpu-installation-us/installer/v1.8.1/cuda_installer.pyz
python3 cuda_installer.pyz install_cuda

Changes

• Updating Secure Boot certificates, as Microsoft has released new set. Read more in Microsoft article and Google Cloud docs.
• Updating driver and CUDA Toolkit versions to 590 and 13.1 for new feature branch.
• Added Debian 13 support.

Requirements

The system on which you want to run the script needs to meet the following
requirements:

• Python interpreter in version 3.8 or newer installed.
• Access to Google Cloud Storage.
• Access to distribution specific repositories, to perform required updates and dependency installations.

Running the tool

The cuda_installer.pyz script needs to be executed with root privileges to install drivers or CUDA Toolkit
(for example sudo python3 cuda_installer.pyz install_cuda).

Note: During the installation the script will trigger system reboots. After a
reboot, the script needs to be started again to continue the installation process.

After successfully installation, the tool will restart your system once more to make
sure everything is initialized properly and working system-wide.

You can use the following script as a startup script for your instance, to make the installation automatic:

#!/bin/bash
if test -f /opt/google/cuda-installer
then
  exit
fi

mkdir -p /opt/google/cuda-installer/
cd /opt/google/cuda-installer/ || exit

curl -fSsL -O https://storage.googleapis.com/compute-gpu-installation-us/installer/v1.8.0/cuda_installer.pyz
python3 cuda_installer.pyz install_cuda

Changes

• Fixing #47 - build_image command won't fail with an error on non-GCP machines. The failing call to metadata.google.internal correctly results now in a False result of the Virtual Workstation test.

Requirements

The system on which you want to run the script needs to meet the following
requirements:

• Python interpreter in version 3.8 or newer installed.
• Access to Google Cloud Storage.
• Access to distribution specific repositories, to perform required updates and dependency installations.

Running the tool

The cuda_installer.pyz script needs to be executed with root privileges to install drivers or CUDA Toolkit
(for example sudo python3 cuda_installer.pyz install_cuda).

Note: During the installation the script will trigger system reboots. After a
reboot, the script needs to be started again to continue the installation process.

After successfully installation, the tool will restart your system once more to make
sure everything is initialized properly and working system-wide.

You can use the following script as a startup script for your instance, to make the installation automatic:

#!/bin/bash
if test -f /opt/google/cuda-installer
then
  exit
fi

mkdir -p /opt/google/cuda-installer/
cd /opt/google/cuda-installer/ || exit

curl -fSsL -O https://storage.googleapis.com/compute-gpu-installation-us/installer/v1.7.1/cuda_installer.pyz
python3 cuda_installer.pyz install_cuda

Changes

• Driver versions upgraded to 580 for LTS and PROD branches. New Feature Branch still at 575.
• RTX Virtual Workstation driver installation support added.

Requirements

The system on which you want to run the script needs to meet the following
requirements:

• Python interpreter in version 3.8 or newer installed.
• Access to Google Cloud Storage.
• Access to distribution specific repositories, to perform required updates and dependency installations.

Running the tool

The cuda_installer.pyz script needs to be executed with root privileges to install drivers or CUDA Toolkit
(for example sudo python3 cuda_installer.pyz install_cuda).

Note: During the installation the script will trigger system reboots. After a
reboot, the script needs to be started again to continue the installation process.

After successfully installation, the tool will restart your system once more to make
sure everything is initialized properly and working system-wide.

You can use the following script as a startup script for your instance, to make the installation automatic:

#!/bin/bash
if test -f /opt/google/cuda-installer
then
  exit
fi

mkdir -p /opt/google/cuda-installer/
cd /opt/google/cuda-installer/ || exit

curl -fSsL -O https://storage.googleapis.com/compute-gpu-installation-us/installer/v1.7.0/cuda_installer.pyz
python3 cuda_installer.pyz install_cuda

Changes

• Introducing new --network and --subnet option for the build_image command. This will allow you to build disk images with signed drivers in projects that don't have the default VPC network (or if you just don't want to use the default VPC network).
• Debian 11 is no longer supported.
• Long Term Support (LTS) driver branch is added as an option for --installation-branch. This installation is possible only in binary mode. To use this branch, use --installation-mode=binary and --installation-branch=lts.

Requirements

The system on which you want to run the script needs to meet the following
requirements:

• Python interpreter in version 3.8 or newer installed.
• Access to Google Cloud Storage.
• Access to distribution specific repositories, to perform required updates and dependency installations.

Running the tool

The cuda_installer.pyz script needs to be executed with root privileges to install drivers or CUDA Toolkit
(for example sudo python3 cuda_installer.pyz install_cuda).

Note: During the installation the script will trigger system reboots. After a
reboot, the script needs to be started again to continue the installation process.

After successfully installation, the tool will restart your system once more to make
sure everything is initialized properly and working system-wide.

You can use the following script as a startup script for your instance, to make the installation automatic:

#!/bin/bash
if test -f /opt/google/cuda-installer
then
  exit
fi

mkdir -p /opt/google/cuda-installer/
cd /opt/google/cuda-installer/ || exit

curl -fSsL -O https://storage.googleapis.com/compute-gpu-installation-us/installer/v1.6.0/cuda_installer.pyz
python3 cuda_installer.pyz install_cuda

Changes

• Introducing new option --installation-branch which controls which NVIDIA driver branch gets installed. The two supported branches are:
  • Production branch (--installation-branch=prod) Installs driver from the 570 series and 12.8 CUDA Toolkit.
  • New Feature branch (--installation-branch=nfb) Installs driver from the 575 series and 12.9 CUDA Toolkit.
  • By default, the tool uses repository installation mode with production branch.
  • Debian 12 does not support the default repository + production branch installation mode. For Debian 12 please use --installation-mode and --installation-branch settings to use a supported mode and branch (binary+prod, repo+nfb or binary+nfb).
  • Debian 11 does not support repository mode at all.

Requirements

The system on which you want to run the script needs to meet the following
requirements:

• Python interpreter in version 3.8 or newer installed.
• Access to Google Cloud Storage.
• Access to distribution specific repositories, to perform required updates and dependency installations.

Running the tool

The cuda_installer.pyz script needs to be executed with root privileges to install drivers or CUDA Toolkit
(for example sudo python3 cuda_installer.pyz install_cuda).

Note: During the installation the script will trigger system reboots. After a
reboot, the script needs to be started again to continue the installation process.

After successfully installation, the tool will restart your system once more to make
sure everything is initialized properly and working system-wide.

You can use the following script as a startup script for your instance, to make the installation automatic:

#!/bin/bash
if test -f /opt/google/cuda-installer
then
  exit
fi

mkdir -p /opt/google/cuda-installer/
cd /opt/google/cuda-installer/ || exit

curl -fSsL -O https://storage.googleapis.com/compute-gpu-installation-us/installer/v1.5.0/cuda_installer.pyz
python3 cuda_installer.pyz install_cuda

Changes

• Updated to 575.51.03 driver and 12.9.0 CUDA Toolkit versions.
• There are now two modes of installation: binary and repo. The repo mode (default) adds NVIDIA repository to the system and downloads the driver and CUDA Toolkit from that repository. binary mode continues to download runfile installers from GCS bucket and installs using those files. The repo mode is best suited for situations where you want to install additional software from the NVIDIA repository, while binary is best when you don't want to have your repository list affected by what's available in NVIDIA repository.
• There is now build_image command that automates the process of creating a Secure Boot-compatible disk image. Find more information about this command with build_image --help or in the README.

Requirements

The system on which you want to run the script needs to meet the following
requirements:

• Python interpreter in version 3.6 or newer installed.
• Access to Google Cloud Storage.
• Access to distribution specific repositories, to perform required updates and dependency installations.

Running the tool

The cuda_installer.pyz script needs to be executed with root privileges to install drivers or CUDA Toolkit
(for example sudo python3 cuda_installer.pyz).

Note: During the installation the script will trigger system reboots. After a
reboot, the script needs to be started again to continue the installation process.

After successfully installation, the tool will restart your system once more to make
sure everything is initialized properly and working system-wide.

You can use the following script as a startup script for your instance, to make the installation automatic:

#!/bin/bash
if test -f /opt/google/cuda-installer
then
  exit
fi

mkdir -p /opt/google/cuda-installer/
cd /opt/google/cuda-installer/ || exit

curl -fSsL -O https://storage.googleapis.com/compute-gpu-installation-us/installer/v1.4.0/cuda_installer.pyz
python3 cuda_installer.pyz install_cuda

Changes

• Updated to 570.124.06 driver and 12.8.1 CUDA Toolkit versions.
• Added the ability to sign installed driver with provided keys. To do that, use --secure-boot-pub-key and --secure-boot-priv-key flags when running install_driver command.
• Added the ability to bypass the requirement for a GPU being connected to the machine. Use --ignore-no-gpu flag for install_driver and install_cuda to use it.
• Switched the source of downloads from NVIDIA to mirrored resources stored in Google Cloud Storage (Google-owned bucket). Now the VMs don't need full access to the Internet to perform the installation + downloads might get faster, as the traffic stays inside Google Cloud network.

Requirements

The system on which you want to run the script needs to meet the following
requirements:

• Python interpreter in version 3.6 or newer installed.
• Access to Google Cloud Storage.
• Access to distribution specific repositories, to perform required updates and dependency installations.

Running the tool

The cuda_installer.pyz script needs to be executed with root privileges
(for example sudo python3 cuda_installer.pyz).

Note: During the installation the script will trigger system reboots. After a
reboot, the script needs to be started again to continue the installation process.

After successfully installation, the tool will restart your system once more to make
sure everything is initialized properly and working system-wide.

You can use the following script as a startup script for your instance, to make the installation automatic:

#!/bin/bash
if test -f /opt/google/cuda-installer
then
  exit
fi

mkdir -p /opt/google/cuda-installer/
cd /opt/google/cuda-installer/ || exit

curl -fSsL -O https://storage.googleapis.com/compute-gpu-installation-us/installer/v1.3.0/cuda_installer.pyz
python3 cuda_installer.pyz install_cuda

Changes

Updated to 560.35.03 driver and 12.6.3 CUDA Toolkit versions.

Requirements

The system on which you want to run the script needs to meet the following
requirements:

• Python interpreter in version 3.6 or newer installed.
• Access to Internet (the script needs to download the driver and CUDA toolkit).
• At least one GPU unit attached.

Running the tool

The cuda_installer.pyz script needs to be executed with root privileges
(for example sudo python3 cuda_installer.pyz).

Note: During the installation the script will trigger system reboots. After a
reboot, the script needs to be started again to continue the installation process.

After successfully installation, the tool will restart your system once more to make
sure everything is initialized properly and working system-wide.

You can use the following script as a startup script for your instance, to make the installation automatic:

#!/bin/bash
if test -f /opt/google/cuda-installer
then
  exit
fi

mkdir -p /opt/google/cuda-installer/
cd /opt/google/cuda-installer/ || exit

curl -fSsL -O https://github.com/GoogleCloudPlatform/compute-gpu-installation/releases/download/cuda-installer-v1.2.0/cuda_installer.pyz
python3 cuda_installer.pyz install_cuda