Operating system details
Stay organized with collections
Save and categorize content based on your preferences.
Understand the operating system (OS) image details, feature support, and
lifecycle stages available on Compute Engine. Use these reference
tables to verify Compute Engine instance compatibility with specific
GPU, networking, security, and import capabilities. Some OS images are
customized specifically to run on Compute Engine and have notable
differences from standard vendor images; these differences are also covered for
each OS.
While these tables cover feature compatibility for Google-provided public OS
image families, Compute Engine also supports private custom images and
specialized public images. For information about custom OS images, image
families, and specialized categories like community-supported, HPC, or TPU
images, see Operating system images.
AlmaLinux
AlmaLinux is an open source, community
owned and governed, forever-free enterprise Linux distribution, focused on long-term stability,
providing a robust production-grade platform. AlmaLinux OS is binary compatible with Red Hat
Enterprise Linux (RHEL). The AlmaLinux OS Foundation builds and supports the AlmaLinux images available for Compute Engine. There is no license fee for
using AlmaLinux with Compute Engine.
Automatic updates
By default, this operating system is configured to install security updates by using the
dnf-automatic
tool. The updates have the following behaviors:
The dnf-automatic tool does not upgrade VMs between
major versions of the operating system.
The upgrade tool is configured to only apply updates marked by the vendor as security updates.
Some updates require reboots to take effect. These reboots do not happen automatically.
Image configuration
AlmaLinux images are built and maintained by the AlmaLinux OS Foundation. AlmaLinux images are
always built with the latest AlmaLinux packages which reflect the most recent AlmaLinux point
release. The AlmaLinux images that are provided by Compute Engine, have the following differences in configuration from standard AlmaLinux cloud images:
Account configuration
There are no local users configured with passwords.
Bootloader configuration
To force faster boot times, the boot timeout in the grub configuration is set to 0.
Network configuration
The SSH server configuration is set up as follows:
Password authentication is disabled.
Root login is disabled.
By default, all traffic is allowed through the guest firewall because the
VPC firewall rules overrides the guest firewall
rules. The guest firewall rules remains enabled and can be configured through normal AlmaLinux methods.
CentOS Linux is
a free operating system that is derived from Red Hat Enterprise Linux (RHEL). Google Cloud
builds and supports the CentOS images available for Compute Engine. There is no license fee for
using CentOS with Compute Engine.
CentOS Stream is a
distribution that is continuously delivered and tracks just ahead of RHEL development.
CentOS Stream is positioned as a midstream development platform between Fedora Linux and RHEL.
Automatic updates
By default, this operating system is configured to install security updates by using the
dnf-automatic
tool. The updates have the following behaviors:
The dnf-automatic tool does not upgrade VMs between
major versions of the operating system.
The upgrade tool is configured to only apply updates marked by the vendor as security updates.
Some updates require reboots to take effect. These reboots do not happen automatically.
Image configuration
The CentOS Stream images that are provided by Compute Engine, have the
following differences in configuration from standard CentOS Stream images:
Account configuration
There are no local users configured with passwords.
Bootloader configuration
To force faster boot times, the boot timeout in the grub configuration is set to 0.
Network configuration
IPv6 is enabled.
The SSH server configuration is set up as follows:
Password authentication is disabled.
To prevent SSH disconnections, ServerAliveInterval and
ClientAliveInterval are set to 7 minutes.
Root login is disabled.
/etc/udev/rules.d/75-persistent-net-generator.rules is disabled.
To prevent MAC addresses from persisting, /etc/udev/rules.d/70-persistent-net.rules
is removed.
By default, all traffic is allowed through the guest firewall because the
VPC firewall rules overrides the guest firewall
rules. The guest firewall rules remains enabled and can be configured through normal CentOS methods.
VMs based on Google-provided Linux images get their interface MTU
from the attached VPC MTU. VMs based on custom images or older
Linux images may have their MTU's hardcoded. In these cases, you have to change
the setting yourself if you want to connect the interface to a network with an
MTU other than 1460. For more information about network and interface MTU, see
the maximum transmission unit overview.
1 If a multiple NIC VM is created using this OS image, the VM
might lose network connectivity after rebooting. This happens if one of these
NICs uses a non-VirtIO interface. For more information, see
known issues.
1 Only available with certain machine series. 2
This OS image has predictable network interface names disabled. Newer image
families may have a different network interface naming scheme.
GPU support
N1+GPU denotes support for NVIDIA T4, V100, or P4 GPUs running on a
general-purpose N1 machine family.
1 If a multiple NIC VM is created using this OS image, the VM
might lose network connectivity after rebooting. This happens if one of these
NICs uses a non-VirtIO interface. For more information, see
known issues.
Security features
OS version
Shielded VM
Confidential VM SEV
Confidential VM SEV-SNP
Confidential VM Intel TDX
CentOS Stream 8
Yes
No
No
No
CentOS 8
Yes
No
No
No
CentOS 7
Yes
No
No
No
CentOS 6
No
No
No
No
1 This OS image doesn't support Secure Boot on ARM64. 2 Intel TDX isn't supported on ARM64 for this OS image.
1 Only available with certain machine series. 2
This OS image has predictable network interface names disabled. Newer image
families may have a different network interface naming scheme.
GPU support
N1+GPU denotes support for NVIDIA T4, V100, or P4 GPUs running on a
general-purpose N1 machine family.
Container-Optimized OS from Google is an operating system image for your
Compute Engine instances that is optimized for running Docker
containers. Google Cloud builds and supports the Container-Optimized OS images available
for Compute Engine. There is no license fee for using Container-Optimized OS with
Compute Engine.
For more information about Container-Optimized OS, see the Container-Optimized OS
overview or
release notes.
Automatic updates
This operating system can be configured to install security updates by using
Automatic updates.
On versions 117 and later, automatic updates are disabled by default. On versions 113 and
earlier, automatic updates are enabled by default.
The automatic updates have the following behaviors:
They don't upgrade instances between major versions of the operating system.
They might require reboots to take effect. These reboots don't happen automatically.
Image configuration
Network configuration
VMs based on Google-provided Linux images get their interface MTU
from the attached VPC MTU. VMs based on custom images or older
Linux images may have their MTU's hardcoded. In these cases, you have to change
the setting yourself if you want to connect the interface to a network with an
MTU other than 1460. For more information about network and interface MTU, see
the maximum transmission unit overview.
GA COS versions
The following section summarizes feature support for versions of this OS that are in General Availability (GA).
N1+GPU denotes support for NVIDIA T4, V100, or P4 GPUs running on a
general-purpose N1 machine family.
For G2 VMs, the current default driver for Container-Optimized OS, don't support L4 GPUs running
on G2 machine types. You might be able to install a supported version, see
G2 limitations.
N1+GPU denotes support for NVIDIA T4, V100, or P4 GPUs running on a
general-purpose N1 machine family.
For G2 VMs, the current default driver for Container-Optimized OS, don't support L4 GPUs running
on G2 machine types. You might be able to install a supported version, see
G2 limitations.
Debian is a free operating system offered by the
Debian community. Google Cloud
builds and supports the Debian images available for Compute Engine. There is no license fee for
using Debian with Compute Engine.
Automatic updates
By default, this operating system is configured to install security updates by using the Debian
UnattendedUpgrades
tool. The updates have the following behaviors:
The UnattendedUpgrades tool does not upgrade VMs between
major versions of the operating system.
The UnattendedUpgrades tool is configured to only automatically apply updates
obtained from the Debian security repository.
Some updates require reboots to take effect. These reboots do not happen automatically.
Debian build tools come from the Debian Cloud team
image project.
Debian images are always built with the latest Debian packages which reflect the most recent
Debian point release.
The Debian images that are provided by Compute Engine, have the following differences in
configuration from standard Debian images:
Account configuration
There are no local users configured with passwords.
Bootloader configuration
To force faster boot times, the boot timeout in the grub configuration is set to 0.
To allow SCSI block multi-queue usage, scsi_mod.use_blk_mq is enabled.
Network configuration
IPv6 is enabled.
The SSH server configuration is set up as follows:
Password authentication is disabled.
Root login is disabled.
To prevent MAC addresses from persisting, /etc/udev/rules.d/70-persistent-net.rules
is removed.
VMs based on Google-provided Linux images get their interface MTU
from the attached VPC MTU. VMs based on custom images or older
Linux images may have their MTU's hardcoded. In these cases, you have to change
the setting yourself if you want to connect the interface to a network with an
MTU other than 1460. For more information about network and interface MTU, see
the maximum transmission unit overview.
Package system and repository configuration
Google Cloud repositories are enabled to install packages for the Compute Engine
guest environment
and the Google Cloud CLI. The guest environment
packages and the Google Cloud CLI packages are installed and enabled by default.
The APT sources are set to use the Debian CDN.
The Unattended-upgrades
package is installed and configured to download and install Debian security updates daily. This
can be configured or disabled by changing the values in /etc/apt/apt.conf.d/50unattended-upgrades
and /etc/apt/apt.conf.d/02periodic.
The cloud-initramfs-growroot package is removed and replaced with
the Google supported
gce-disk-expand
package.
The linux-image-cloud-amd64 kernel is installed instead of the generic Debian kernel.
The haveged package is installed to provide entropy.
(Debian 13 only) The gce-configs-trixie package is installed to enable
VM serial port logging and ssh-in-browser, which requires additional configuration due to systemd
updates in Debian 13.
(Debian 13 only) The gce-configs-trixie package also provides the
google-keyring.gpg key installed to /etc/apt/keyrings to comply with
updated repository authentication guidance. The
/etc/apt/sources.list.d/google-cloud.list file uses this key to authenticate against
Google Cloud repositories.
Storage configuration
Images are 10 GB by default.
The partition table is GPT, and there is an EFI partition to
support booting on UEFI. There is also an MBR boot block to support BIOS.
The floppy module is disabled because there is no floppy disk controller on Compute Engine.
Time configuration
The NTP server is set to use the Compute Engine metadata server.
GA Debian versions
The following section summarizes feature support for versions of this OS that are in General Availability (GA).
1 Only available with certain machine series. 2 Fully supported with VirtIO, but requires an updated driver to use with gVNIC.
For more information, see
Jumbo frames. 3 You can update the gVNIC driver
to the latest version to enable network egress bandwidths of 200 Gbps.
For more information, see the
Requirements and limitations
section of "Configure per VM Tier_1 networking performance".
GPU support
N1+GPU denotes support for NVIDIA T4, V100, or P4 GPUs running on a
general-purpose N1 machine family.
1 Only available with certain machine series. 2 Fully supported with VirtIO, but requires an updated driver to use with gVNIC.
For more information, see
Jumbo frames. 3 You can update the gVNIC driver
to the latest version to enable network egress bandwidths of 200 Gbps.
For more information, see the
Requirements and limitations
section of "Configure per VM Tier_1 networking performance".
GPU support
N1+GPU denotes support for NVIDIA T4, V100, or P4 GPUs running on a
general-purpose N1 machine family.
Fedora CoreOS is a distribution that provides features that are needed to run modern
infrastructure stacks. Fedora CoreOS uses Linux containers to manage your services
at a higher level of abstraction. Google Cloud provides Fedora CoreOS
images built and supported by Fedora. There is no license fee for using Fedora CoreOS
with Compute Engine.
Automatic updates
By default, this operating system is configured to install security updates by using the FedoraCoreOS
automatic update
tool. The updates have the following behaviors:
These automatic updates from the operating system vendor do not upgrade instances between
major versions of the operating system.
Some updates require reboots to take effect. These reboots do not happen automatically.
Image configuration
Network configuration
VMs based on Google-provided Linux images get their interface MTU
from the attached VPC MTU. VMs based on custom images or older
Linux images may have their MTU's hardcoded. In these cases, you have to change
the setting yourself if you want to connect the interface to a network with an
MTU other than 1460. For more information about network and interface MTU, see
the maximum transmission unit overview.
GA Fedora CoreOS versions
The following section summarizes feature support for versions of this OS that are in General Availability (GA).
Oracle Linux
is a Linux operating system (OS) provided by Oracle with everything required to deploy, optimize,
and manage applications on-premises, in the cloud, and at the edge.
Oracle builds and supports the Oracle Linux images available for Compute Engine. There is no
license fee for using Oracle Linux with Compute Engine. Oracle is responsible for ensuring that
Oracle Linux works with Google Cloud features and that security updates are maintained. For
issues related to Oracle Linux, you must consult community resources or get enterprise-level
support directly from Oracle.
dnf-automatic doesn't upgrade VMs between major versions of the operating
system.
The upgrade tool only applies updates marked by Oracle as security updates.
Some updates require reboots to take effect. These reboots don't happen
automatically.
Image configuration
The Oracle Linux images that are provided by Compute Engine reflect the
most recent point release. Currently you cannot pin a VM to a point release. These
images have the following differences in configuration from standard Oracle Linux
images:
Account configuration
There are no local users configured with passwords.
Bootloader configuration
To force faster boot times, the boot timeout in the grub configuration is set to 0.
The I/O scheduler is set to noop.
Network configuration
IPv6 is enabled.
The SSH server configuration is set up as follows:
Password authentication is disabled.
To prevent SSH disconnections, ServerAliveIntervaland
ClientAliveInterval are set to 7 minutes.
Root login is disabled.
/etc/udev/rules.d/75-persistent-net-generator.rules is disabled.
To prevent MAC addresses from persisting, /etc/udev/rules.d/70-persistent-net.rules
is removed.
By default, all traffic is allowed through the guest firewall because the
VPC firewall rules overrides the guest firewall
rules. The guest firewall rules remains enabled and can be configured through normal Oracle Linux methods.
VMs based on Google-provided Linux images get their interface MTU
from the attached VPC MTU. VMs based on custom images or older
Linux images may have their MTU's hardcoded. In these cases, you have to change
the setting yourself if you want to connect the interface to a network with an
MTU other than 1460. For more information about network and interface MTU, see
the maximum transmission unit overview.
Package system and repository configuration
The Google Cloud repository is enabled to install packages for the Compute Engine
guest environment and disabled for the
Google Cloud CLI. The Google Cloud CLI
can be installed as follows:
dnf install -y google-cloud-cli --enablerepo google-cloud-sdk
Automatic security updates are disabled by default. To enable, refer to the
instructions in the Automatic updates section.
Storage configuration
By default, images are 20 GB. This is the recommended minimum size.
The partition table is GPT, and there is an EFI
partition to support booting on UEFI.
Time configuration
The NTP server is set to use the Compute Engine metadata server.
GA Oracle Linux versions
The following section summarizes feature support for versions of this OS that are in General Availability (GA).
Red Hat Enterprise Linux
(RHEL) is an open-source Linux operating system that provides both server
and desktop operating systems. Google Cloud builds and supports the RHEL OS images available for
Compute Engine, including select images pre-configured with Logical Volume Manager (LVM) partitioning.
RHEL images are
premium resources that incur
additional fees to use.
If you want to use an existing RHEL subscription, you can use the
Red Hat Cloud Access feature.
The Red Hat Knowledgebase provides you with access to articles, solutions, product documentation,
and community discussions. The Red Hat Knowledgebase
is available as a single-sign-on (SSO) option through the Google Cloud console. See
Access Red Hat Knowledgebase.
Red Hat provides Extended Update Support (EUS) for specific minor releases of RHEL. EUS extends
the support duration for these point releases, offering security updates and critical fixes for
up to 2 years after the initial release. For more information, see the Red Hat EUS Overview.
The RHEL 10.2 EUS image has an additional offering containing a out-of-tree (OOT) Google Virtual
Network Interface Card (gVNIC) driver. This offering is intended only for the RHEL
10.2 EUS release. The gVNIC driver will be rolled upstream before the next EUS release (RHEL 10.4)
and be a part of all subsequent releases.
To view a list of frequently asked questions when running RHEL on Compute Engine, see
Red Hat Enterprise Linux FAQ.
Automatic updates
By default, this operating system is configured to install security updates by using the RHEL
yum-cron (RHEL 7) or dnf automatic (RHEL 8+) tool. The updates have the
following behaviors:
These automatic updates from the operating system vendor do not upgrade instances between
major versions of the operating system.
Starting with RHEL 7, the operating system is also configured to only apply updates
marked by the vendor as security updates.
Some updates require reboots to take effect. These reboots do not happen automatically.
RHEL images are always built with the latest RHEL packages, which reflect the most recent point release.
Currently, you cannot pin a VM to a point release.
RHEL for SAP images are tagged to the specific point release they are built for as supported by Red Hat.
The RHEL images that are provided by Compute Engine, have the following differences in
configuration from standard RHEL images:
Account configuration
There are no local users configured with passwords.
Bootloader configuration
To force faster boot times, the boot timeout in the grub configuration is set to 0.
The I/O scheduler is set to noop.
EUS configuration
Before the RHEL EUS version becomes EOL, it will need to be upgraded to the EUS version. You can
check your current EUS version by running the following command:
cat /etc/dnf/vars/releasever
To lock the RHEL to the new point release version, run the following command:
echo \"VERSION\" | sudo tee /etc/dnf/vars/releasever
Replace VERSION with the version in the format of major.minor.
For example, for version 9.6 specify 9.6.
To upgrade to the next RHEL EUS version, run the following command:
sudo dnf -y --releasever=VERSION upgrade
Replace VERSION with the version in the format of major.minor.
For example, for version 9.6 specify 9.6.
Logical Volume Manager (LVM) configuration
LVM-enabled RHEL images include a predefined partitioning scheme:
EFI system partition: 200 MB for bootloader files.
/boot partition: 1 GB (XFS filesystem) for kernel images and boot-related
files.
LVM Physical Volume: A single LVM Physical Volume (PV) uses most of the disk space. This PV belongs to a Volume Group named rootvg.
Within the rootvg Volume Group, the following Logical Volumes (LVs) are
created by default:
LV name
Mount point
Default size
Filesystem
rootlv
/
2.0 GB
XFS
usrlv
/usr
10.0 GB
XFS
varlv
/var
8.0 GB
XFS
tmplv
/tmp
2.0 GB
XFS
homelv
/home
1.0 GB
XFS
Unallocated space: The remaining space on the disk within rootvg is left
unallocated. This lets you create new Logical Volumes or extend the
existing ones as needed. The minimum recommended disk size for these images
is 50GB.
Key differences from standard RHEL images:
No automatic disk expansion: Unlike standard Compute Engine RHEL images,
the LVM images do not include the gce-disk-expand package. When you
resize the boot disk in the Google Cloud console or by using gcloud, the
underlying Physical Volume and Volume Group see the additional space,
but the Logical Volumes (such as rootlv or usrlv) don't
automatically grow. You must use standard LVM commands to use the new
space.
LVM tools: The lvm2 package is installed by default,
providing all necessary command-line tools (pvcreate, vgcreate,
lvcreate, lvextend, resize2fs, xfs_growfs, etc.).
Filesystem layout: The root filesystem and other system directories like
/usr, /var are on separate Logical Volumes, as detailed in the Logical Volumes table.
Network configuration
IPv6 is enabled.
The SSH server configuration is set up as follows:
Password authentication is disabled.
To prevent SSH disconnections, ServerAliveInterval and
ClientAliveInterval are set to 7 minutes.
Root login is disabled.
/etc/udev/rules.d/75-persistent-net-generator.rules is disabled.
To prevent MAC addresses from persisting, /etc/udev/rules.d/70-persistent-net.rules
is removed.
By default, all traffic is allowed through the guest firewall because the
VPC firewall rules overrides the guest firewall
rules. The guest firewall rules remains enabled and can be configured through normal RHEL methods.
VMs based on Google-provided Linux images get their interface MTU
from the attached VPC MTU. VMs based on custom images or older
Linux images may have their MTU's hardcoded. In these cases, you have to change
the setting yourself if you want to connect the interface to a network with an
MTU other than 1460. For more information about network and interface MTU, see
the maximum transmission unit overview.
RHEL for SAP yum vars are set to version lock the client to the supported RHEL for SAP point release.
Compute Engine offers Red Hat Update Infrastructure (RHUI) servers for on-demand
RHEL images. If your VMs don't have external IP addresses assigned, they can
use Private Google Access. This access lets them reach the RHUI servers at 35.190.247.88
and 2600:1901:0:202d:: for package updates. You must ensure that
firewall rules permit egress to these IP addresses. For networks that don't
have a default route to the internet, you must create a custom route for
destination 35.190.247.88/32 to the default internet
gateway next hop. For more information about Private Google Access, see Private Google Access.
If you're unable to get updates
from the Red Hat Update Infrastructure (RHUI) servers, the RHUI client package might need to be
updated. To update to the latest RHUI client package, run the following command:
dnf -y -q update 'google-rhui-client*'
The Google RHUI client package, which contains the configuration needed
to access RHEL content, is installed.
The Red Hat subscription-manager package is removed because it is not used for pay as you go
images.
Automatic updates are enabled as follows:
For RHEL 8+, by using dnf automatic.
For all versions, the update_cmd property is set to security.
Storage configuration
By default, images are 20 GB. This is the recommended minimum size.
The partition table is GPT, and there is an EFI
partition to support booting on UEFI.
Time configuration
The NTP server is set to use the Compute Engine metadata server.
GA RHEL versions
The following section summarizes feature support for versions of this OS that are in General Availability (GA).
2ELS: To use this OS image, you must have a subscription or an Extended Life
Cycle Support (ELS) Add-On. You can append the ELS Add-On to your RHEL compute instances as
described in
Append RHEL ELS licenses.
3LVM: Logical Volume Manager (LVM) is a disk management tool that abstracts
physical storage into volume groups, allowing you to span file systems across multiple disks and
resize them dynamically. For detailed information, see the official
Red Hat LVM documentation.
1This OS image supports NVMe but does not include all optimizations for NVMe.
2If a compute instance withmultiple NICs is created using this OS image, the instance
might lose network connectivity after rebooting. This happens if one of these NICs uses a
non-VirtIO interface. See
known issues.
1 Only available with certain machine series. 2
This OS image has predictable network interface names disabled. Newer image
families may have a different network interface naming scheme.
3 You can
update the gVNIC driver to
the latest version to enable network egress bandwidths of 200 Gbps. For more information, see the
Requirements and limitations
section of "Configure per VM Tier_1 networking performance". 4 Fully supported with VirtIO, but requires an updated driver to use with gVNIC. For
more information, see Jumbo frames.
GPU support
N1+GPU denotes support for NVIDIA T4, V100, or P4 GPUs running on a
general-purpose N1 machine family.
1VM Manager provides vulnerability scanning results for RHEL based on the
latest minor version for each major version released. If your compute instance runs an older
minor version of RHEL, then you might get inaccurate results in the vulnerability reports.
Import
For OS support information on migrating compute instances using Migrate to Virtual Machines, see
supported operating systems.
Extended Update Support (EUS) is an offering that can be added to your Red Hat Enterprise
Linux (RHEL) subscription. For this reason, each RHEL EUS image includes two licenses:
1ELS: To use this OS image, you must have a subscription or an Extended Life
Cycle Support (ELS) Add-On. You can append the ELS Add-On to your compute instances as described
in Append RHEL ELS licenses.
2LVM: This image supports Logical Volume Manager (LVM) partitioning.
If you use LVM, your image also includes the following license:
https://www.googleapis.com/compute/v1/projects/rhel-cloud/global/licenses/rhel-lvm.
End of support RHEL versions
The following section summarizes feature support for versions of this OS that are in end of support.
2ELS: To use this OS image, you must have a subscription or an Extended Life
Cycle Support (ELS) Add-On. You can append the ELS Add-On to your RHEL compute instances as
described in
Append RHEL ELS licenses.
3LVM: Logical Volume Manager (LVM) is a disk management tool that abstracts
physical storage into volume groups, allowing you to span file systems across multiple disks and
resize them dynamically. For detailed information, see the official
Red Hat LVM documentation.
1This OS image supports NVMe but does not include all optimizations for NVMe.
2If a compute instance withmultiple NICs is created using this OS image, the instance
might lose network connectivity after rebooting. This happens if one of these NICs uses a
non-VirtIO interface. See
known issues.
1 Only available with certain machine series. 2
This OS image has predictable network interface names disabled. Newer image
families may have a different network interface naming scheme.
3 You can
update the gVNIC driver to
the latest version to enable network egress bandwidths of 200 Gbps. For more information, see the
Requirements and limitations
section of "Configure per VM Tier_1 networking performance". 4 Fully supported with VirtIO, but requires an updated driver to use with gVNIC. For
more information, see Jumbo frames.
GPU support
N1+GPU denotes support for NVIDIA T4, V100, or P4 GPUs running on a
general-purpose N1 machine family.
1VM Manager provides vulnerability scanning results for RHEL based on the
latest minor version for each major version released. If your compute instance runs an older
minor version of RHEL, then you might get inaccurate results in the vulnerability reports.
Import
For OS support information on migrating compute instances using Migrate to Virtual Machines, see
supported operating systems.
Extended Update Support (EUS) is an offering that can be added to your Red Hat Enterprise
Linux (RHEL) subscription. For this reason, each RHEL EUS image includes two licenses:
1ELS: To use this OS image, you must have a subscription or an Extended Life
Cycle Support (ELS) Add-On. You can append the ELS Add-On to your compute instances as described
in Append RHEL ELS licenses.
2LVM: This image supports Logical Volume Manager (LVM) partitioning.
If you use LVM, your image also includes the following license:
https://www.googleapis.com/compute/v1/projects/rhel-cloud/global/licenses/rhel-lvm.
Rocky Linux
Rocky Linux is a free, open,
community enterprise operating system designed to be 100% bug-for-bug compatible with Red Hat
Enterprise Linux. Google Cloud builds and supports the Rocky Linux images available for
Compute Engine. There is no license fee for using Rocky Linux with Compute Engine.
The following three versions of Rocky Linux operating systems are available on Compute Engine:
A fully open source version
A version optimized for Google Cloud: this version has the suffix -optimized-gcp
and is pre-configured to use the latest version of the Google virtual network interface (gVNIC).
An accelerated version for GPU workloads: this version has the suffix
-optimized-gcp-nvidia-* which is built off of the -optimized-gcp
version and includes the indicated version of the NVIDIA driver.
Automatic updates
By default, this operating system is configured to install security updates by using the
dnf-automatic
tool. The updates have the following behaviors:
dnf-automatic does not upgrade VMs between
major versions of the operating system.
The upgrade tool is configured to only apply updates marked by the vendor as security updates.
Some updates require reboots to take effect. These reboots do not happen automatically.
Image configuration
Rocky Linux images are built and maintained by
CIQ. Rocky Linux images are
always built with the latest Rocky Linux packages which reflect the most recent Rocky Linux
point release.
The Rocky Linux images that are provided by Compute Engine, have the
following differences in configuration from standard Rocky Linux images:
Account configuration
There are no local users configured with passwords.
Bootloader configuration
To force faster boot times, the boot timeout in the grub configuration is set to 0.
Network configuration
IPv6 is enabled.
The SSH server configuration is set up as follows:
Password authentication is disabled.
To prevent SSH disconnections, ServerAliveIntervaland
ClientAliveInterval are set to 7 minutes.
Root login is disabled.
/etc/udev/rules.d/75-persistent-net-generator.rules is disabled.
To prevent MAC addresses from persisting, /etc/udev/rules.d/70-persistent-net.rules
is removed.
By default, all traffic is allowed through the guest firewall because the
VPC firewall rules overrides the guest firewall
rules. The guest firewall rules remains enabled and can be configured through normal Rocky Linux methods.
VMs based on Google-provided Linux images get their interface MTU
from the attached VPC MTU. VMs based on custom images or older
Linux images may have their MTU's hardcoded. In these cases, you have to change
the setting yourself if you want to connect the interface to a network with an
MTU other than 1460. For more information about network and interface MTU, see
the maximum transmission unit overview.
For the Google Cloud optimized and accelerated versions of the Rocky Linux images, the GitLab
CIQ SIG/Cloud Next repository is
enabled. This repository provides a Linux kernel version specifically designed to work best with
Google Cloud hardware.
For the accelerated version of the Rocky Linux image, the GitLab
CIQ SIG/Cloud Next Nonfree repository
is also enabled. This repository provides the proprietary drivers necessary for the system to
fully utilize the GPU hardware.
Repositories are set to use the Rocky Linux default mirror network.
The PowerTools repository is enabled.
Automatic updates are configured as follows:
Automatic updates are enabled by using dnf automatic.
For all versions, the update_cmd property is set to security. However,
by default Rocky Linux does not offer security tagged repositories.
Storage configuration
By default, images are 20 GB.
The partition table is GPT, and there is an EFI
partition to support booting on UEFI.
The floppy module is disabled because there is no floppy disk controller on Compute Engine .
Time configuration
The NTP server is set to use the Compute Engine metadata server.
64k page size kernel configuration for ARM
CIQ SIG/Cloud Next now provides a 64k page size kernel kernel-64k for ARM architecture,
along with kernel modules (kmods) built specifically for it. If you install kernel-64k on
a system with existing SIG/Cloud Next (SCN) kmods, the corresponding 64k-compatible kmods are
automatically installed alongside them. After rebooting into the 64k kernel, the kmods are available
and functional.
This kernel is validated on the C4A machine series with the
kmod-idpf-irdma kernel module, which provides an updated IDPF driver for HPC workloads.
How to enable the 64k page size kernel
Before you enable the 64k kernel, ensure that the CIQ SIG/Cloud Next repository is enabled (set to 1) at the default priority -1 by running the following command:
Non-default kernels, including kernel-64k, undergo less kmod testing than the default kernel. You should validate your specific workloads on this kernel.
GA Rocky Linux versions
The following section summarizes feature support for versions of this OS that are in General Availability (GA).
1 Only available with certain machine series. 2
This OS image has predictable network interface names disabled. Newer image
families may have a different network interface naming scheme.
GPU support
N1+GPU denotes support for NVIDIA T4, V100, or P4 GPUs running on a
general-purpose N1 machine family.
The Rocky Linux 8 and 9 Accelerated OS images are optimized for supporting your
artificial intelligence (AI) and machine learning (ML) workloads. To learn more about how these
images are optimized for these workloads, see
Operating systems in the
AI Hypercomputer documentation.
For tracking purposes, Rocky Linux images include one or more license strings. While most Rocky
Linux images include only a single license string, Rocky Linux Accelerated images include the
following three licenses:
A license for the base Rocky Linux optimized OS.
A license to designate the image as an accelerated image.
1 Only available with certain machine series. 2
This OS image has predictable network interface names disabled. Newer image
families may have a different network interface naming scheme.
GPU support
N1+GPU denotes support for NVIDIA T4, V100, or P4 GPUs running on a
general-purpose N1 machine family.
The Rocky Linux 8 and 9 Accelerated OS images are optimized for supporting your
artificial intelligence (AI) and machine learning (ML) workloads. To learn more about how these
images are optimized for these workloads, see
Operating systems in the
AI Hypercomputer documentation.
For tracking purposes, Rocky Linux images include one or more license strings. While most Rocky
Linux images include only a single license string, Rocky Linux Accelerated images include the
following three licenses:
A license for the base Rocky Linux optimized OS.
A license to designate the image as an accelerated image.
Microsoft SQL Server can be used on Linux-based VMs or images with an on-demand license.
Google does not provide pre-configured images for SQL Server on Linux.
Windows images provided by Google have a hardcoded MTU. For more information about network and
interface MTU, see the maximum transmission unit overview.
SUSE Linux Enterprise Server (SLES),
a versatile server operating system for deploying highly available
enterprise-class IT services in mixed IT environments with improved
performance and reduced risk.
SUSE builds and supports the SUSE images available for Compute Engine. SUSE images
are premium resources that incur
additional fees to use.
This operating system is not configured to install updates by default. For more information about
configuring automatic updates for SLES, see
SUSE documentation.
Image configuration
SLES and SLES for SAP images are built and maintained by SUSE.
SLES images are built with the latest SLES packages reflected in their release.
The SUSE images that are provided by Compute Engine, have the following differences in
configuration from standard SUSE images:
Account configuration
There are no local users configured with passwords.
Network configuration
IPv6 is enabled.
The SSH server configuration is set to disable password authentication.
SLES does not use predictive network interface naming. In the grub kernel command-line arguments,
net.ifnames=0 is set. Therefore, network interfaces use the traditional ethN naming,
with the default interface always being eth0.
VMs based on Google-provided Linux images get their interface MTU
from the attached VPC MTU. VMs based on custom images or older
Linux images may have their MTU's hardcoded. In these cases, you have to change
the setting yourself if you want to connect the interface to a network with an
MTU other than 1460. For more information about network and interface MTU, see
the maximum transmission unit overview.
SLES instances register with a SUSE run SMT service for Compute Engine and are
configured to use SUSE regional mirrors in Compute Engine.
Storage configuration
Images are 10 GB by default.
The partition table is GPT, and there is an EFI partition to support
booting on UEFI. There is also an MBR boot block to support BIOS.
The floppy module is disabled because there is no floppy disk controller on Compute Engine.
Starting with SLES 16, the default file system for the root partition (/) is Btrfs, changing from the previous default of XFS. For more information, see
SUSE file system documentation.
Time configuration
The NTP server is set to use the Compute Engine metadata server.
GA SLES versions
The following section summarizes feature support for versions of this OS that are in General Availability (GA).
1BYOS with LTSS: Support for this operating system is only offered through the
Long Term Service Pack Support (LTSS)
that is available when using BYOS licenses from SUSE.
2You can
update the gVNIC driver to
the latest version to enable network egress bandwidths of 200 Gbps. For more information, see the
Requirements and limitations
section of "Configure per VM Tier_1 networking performance".
3Fully supported with VirtIO, but requires an updated driver to use with gVNIC. For
more information, see Jumbo frames.
GPU support
N1+GPU denotes support for NVIDIA T4, V100, or P4 GPUs running on a general-purpose N1
machine family.
1BYOS with LTSS: Support for this operating system is only offered through the
Long Term Service Pack Support (LTSS)
that is available when using BYOS licenses from SUSE.
End of support SLES versions
The following section summarizes feature support for versions of this OS that are in end of support.
1BYOS with LTSS: Support for this operating system is only offered through the
Long Term Service Pack Support (LTSS)
that is available when using BYOS licenses from SUSE.
2You can
update the gVNIC driver to
the latest version to enable network egress bandwidths of 200 Gbps. For more information, see the
Requirements and limitations
section of "Configure per VM Tier_1 networking performance".
3Fully supported with VirtIO, but requires an updated driver to use with gVNIC. For
more information, see Jumbo frames.
GPU support
N1+GPU denotes support for NVIDIA T4, V100, or P4 GPUs running on a general-purpose N1
machine family.
1BYOS with LTSS: Support for this operating system is only offered through the
Long Term Service Pack Support (LTSS)
that is available when using BYOS licenses from SUSE.
Ubuntu LTS
Ubuntu is a free operating system developed and supported by
Canonical.
Ubuntu Long Term Support (LTS)
images receive bug fixes and security updates for five years after their
release date. LTS images can run on your instances for several years
without having to upgrade to a newer release.
Ubuntu Minimal
images are supported the same as Ubuntu LTS images.
Regular (non LTS) Ubuntu images are supported for 9 months from their
release date. To continue to use a regular Ubuntu image, you will have
to upgrade to the next regular Ubuntu release or LTS release after the
support cycle ends to receive fixes and updates. Compute Engine recommends
using Ubuntu LTS images unless you require features or software packages
that are not yet included in an LTS release. If your instances run Ubuntu
releases that are no longer supported,
upgrade to a supported Ubuntu release.
Automatic updates
By default, this operating system is configured to install security updates by using the Ubuntu
AutomaticSecurityUpdates
tool. The updates have the following behaviors:
The AutomaticSecurityUpdates tool does not upgrade VMs between
major versions of the operating system.
The AutomaticSecurityUpdates tool is configured to only automatically apply
updates obtained from the Ubuntu security repository.
Some updates require reboots to take effect. These reboots do not happen automatically.
Image configuration
Ubuntu images are built and maintained by Canonical. Ubuntu images are always built with the latest
Ubuntu packages which reflect the most recent Ubuntu point release.
The Ubuntu images that are provided by Compute Engine, have the following differences in
configuration from standard Ubuntu images:
Account configuration
There are no local users configured with passwords.
Bootloader configuration
Ubuntu uses cloud-init to do some boot time initialization. The
cloud.cfg file is configured for Compute Engine and enables only the
cloud-init modules that are used.
To force faster boot times, the boot timeout in the grub configuration is set to 0.
To allow SCSI block multi-queue usage, scsi_mod.use_blk_mq is enabled.
Network configuration
IPv6 is enabled.
The SSH server configuration is set to disable password authentication.
To prevent MAC addresses from persisting,
/etc/udev/rules.d/70-persistent-net.rules is removed.
For Ubuntu 18.04+, the Google Cloud CLI is
installed and maintained as a snap package.
The APT sources are set to use the Ubuntu Compute Engine mirrors via cloud-init.
The Unattended-upgrades
package is installed and configured to download and install Debian security updates daily. This
can be configured or disabled by changing the values in /etc/apt/apt.conf.d/50unattended-upgrades
and /etc/apt/apt.conf.d/02periodic.
The linux-image-gcp kernel is used instead of the generic Ubuntu kernel.
The Google Cloud kernel reflects the latest rolling
HWE kernel for Ubuntu LTS.
Storage configuration
Images are 10 GB by default.
The partition table is GPT, and there is an EFI partition to support
booting on UEFI. There is also an MBR boot block to support BIOS.
The floppy module is disabled because there is no floppy disk controller on Compute Engine.
Time configuration
The NTP server is set to use the Compute Engine metadata server.
GA Ubuntu LTS versions
The following section summarizes feature support for versions of this OS that are in General Availability (GA).
1Ubuntu ESM: You can apply your existing
ESM subscription to the Google provided
OS image. The image that is provided by Google Cloud contains enhancements
that might not be included if you bring your own OS image. 2Ubuntu Pro: To continue using Ubuntu 16.04
and 18.04 LTS images,
upgrade from Ubuntu to Ubuntu Pro.
1 Only available with certain machine series. 2 You can
update the gVNIC driver
to the latest version to enable network egress bandwidths of 200 Gbps.
For more information, see the
Requirements and limitations
section of "Configure per VM Tier_1 networking performance". 3 Fully supported with VirtIO, but requires an updated driver to
use with gVNIC. For more information, see
Jumbo frames.
GPU support
N1+GPU denotes support for NVIDIA T4, V100, or P4 GPUs running on a
general-purpose N1 machine family.
The Ubuntu 22.04 and 24.04 LTS Accelerated OS images are optimized for supporting your
artificial intelligence (AI) and machine learning (ML) workloads. To learn more about how these
images are optimized for these workloads, see
Accelerator OS images in the
AI Hypercomputer documentation.
1Ubuntu ESM: You can apply your existing
ESM subscription to the Google provided
OS image. The image that is provided by Google Cloud contains enhancements
that might not be included if you bring your own OS image. 2Ubuntu Pro: To continue using Ubuntu 16.04
and 18.04 LTS images, you must
upgrade from Ubuntu to Ubuntu Pro.
End of support Ubuntu LTS versions
The following section summarizes feature support for versions of this OS that are in end of support.
1Ubuntu ESM: You can apply your existing
ESM subscription to the Google provided
OS image. The image that is provided by Google Cloud contains enhancements
that might not be included if you bring your own OS image. 2Ubuntu Pro: To continue using Ubuntu 16.04
and 18.04 LTS images,
upgrade from Ubuntu to Ubuntu Pro.
1 Only available with certain machine series. 2 You can
update the gVNIC driver
to the latest version to enable network egress bandwidths of 200 Gbps.
For more information, see the
Requirements and limitations
section of "Configure per VM Tier_1 networking performance". 3 Fully supported with VirtIO, but requires an updated driver to
use with gVNIC. For more information, see
Jumbo frames.
GPU support
N1+GPU denotes support for NVIDIA T4, V100, or P4 GPUs running on a
general-purpose N1 machine family.
The Ubuntu 22.04 and 24.04 LTS Accelerated OS images are optimized for supporting your
artificial intelligence (AI) and machine learning (ML) workloads. To learn more about how these
images are optimized for these workloads, see
Accelerator OS images in the
AI Hypercomputer documentation.
1Ubuntu ESM: You can apply your existing
ESM subscription to the Google provided
OS image. The image that is provided by Google Cloud contains enhancements
that might not be included if you bring your own OS image. 2Ubuntu Pro: To continue using Ubuntu 16.04
and 18.04 LTS images, you must
upgrade from Ubuntu to Ubuntu Pro.
Ubuntu Pro
Ubuntu Pro is a premium operating system developed and supported by
Canonical.
Ubuntu Pro
images running on Compute Engine receive bug fixes and security updates for
10 years after their release date, along with access to
Ubuntu ESM. For Ubuntu 16.04,
security updates and ESM are available for 8 years.
Ubuntu Pro images are premium resources that incur additional fees to use.
For information about Ubuntu Pro license cost, see the
Ubuntu Pro pricing page.
An Ubuntu Pro image running on Compute Engine has an on-demand license and does not
require an additional Ubuntu Pro subscription.
To view a list of frequently asked questions when running Ubuntu Pro on Compute Engine, see
Ubuntu Pro FAQ.
Automatic updates
By default, this operating system is configured to install security updates by using the Ubuntu
AutomaticSecurityUpdates
tool. The updates have the following behaviors:
The AutomaticSecurityUpdates tool does not upgrade VMs between
major versions of the operating system.
The AutomaticSecurityUpdates tool is configured to only automatically apply
updates obtained from the Ubuntu security repository.
Some updates require reboots to take effect. These reboots do not happen automatically.
Image configuration
Ubuntu Pro images are built and maintained by Canonical. Ubuntu Pro images are always built with the latest
Ubuntu packages which reflect the most recent Ubuntu point release.
The Ubuntu Pro images that are provided by Compute Engine, have the following differences in
configuration from standard Ubuntu images:
Account configuration
There are no local users configured with passwords.
Bootloader configuration
Ubuntu uses cloud-init to do some boot time initialization. The
cloud.cfg file is configured for Compute Engine and enables only the
cloud-init modules that are used.
To force faster boot times, the boot timeout in the grub configuration is set to 0.
To allow SCSI block multi-queue usage, scsi_mod.use_blk_mq is enabled.
Network configuration
IPv6 is enabled.
The SSH server configuration is set to disable password authentication.
To prevent MAC addresses from persisting,
/etc/udev/rules.d/70-persistent-net.rules is removed.
The Google Cloud CLI is
installed and maintained as a snap package.
The APT sources are set to use the Ubuntu Compute Engine mirrors via cloud-init.
The Unattended-upgrades
package is installed and configured to download and install Debian security updates daily. This
can be configured or disabled by changing the values in /etc/apt/apt.conf.d/50unattended-upgrades
and /etc/apt/apt.conf.d/02periodic.
The linux-image-gcp kernel is used instead of the generic Ubuntu kernel.
The Google Cloud kernel reflects the latest rolling
HWE kernel for Ubuntu LTS.
Storage configuration
Images are 10 GB by default.
The partition table is GPT, and there is an EFI partition to support
booting on UEFI. There is also an MBR boot block to support BIOS.
Ubuntu uses cloud-init to do some boot time initialization. The
cloud.cfg file is configured for Compute Engine and enables only the
cloud-init modules that are used.
There are no local users configured with passwords.
The NTP server is set to use the Compute Engine metadata server.
The floppy module is disabled because there is no floppy disk controller on Compute Engine.
Time configuration
The NTP server is set to use the Compute Engine metadata server.
GA Ubuntu Pro versions
The following section summarizes feature support for versions of this OS that are in General Availability (GA).
1This OS image supports NVMe but doesn't support the hot-plugging of NVMe drives.
Hot-plugging is the process of adding or removing a device while the system is running.
1 Only available with certain machine series. 2 You can
update the gVNIC driver
to the latest version to enable network egress bandwidths of 200 Gbps.
For more information, see the
Requirements and limitations
section of "Configure per VM Tier_1 networking performance". 3 Fully supported with VirtIO, but requires an updated driver to
use with gVNIC. For more information, see
Jumbo frames.
GPU support
N1+GPU denotes support for NVIDIA T4, V100, or P4 GPUs running on a
general-purpose N1 machine family.
1This OS image supports NVMe but doesn't support the hot-plugging of NVMe drives.
Hot-plugging is the process of adding or removing a device while the system is running.
1 Only available with certain machine series. 2 You can
update the gVNIC driver
to the latest version to enable network egress bandwidths of 200 Gbps.
For more information, see the
Requirements and limitations
section of "Configure per VM Tier_1 networking performance". 3 Fully supported with VirtIO, but requires an updated driver to
use with gVNIC. For more information, see
Jumbo frames.
GPU support
N1+GPU denotes support for NVIDIA T4, V100, or P4 GPUs running on a
general-purpose N1 machine family.
The Windows client update settings determine how versions of Windows client use Windows Updates.
To configure Windows automatic updates, see
Configure Automatic Updates.
Storage
Google doesn't test the
VSS agent
on any client images.
GA Windows client versions
The following section summarizes feature support for versions of this OS that are in General Availability (GA).
1Lifecycle stage: Google Cloud Support typically coincides with Extended Support End
Date. For information about the end of support (EOS) date for each of the operating systems in the
table below, see the
Microsoft Lifecycle Policy.
2Windows 11: vTPM and Secure Boot must be enabled. Due to CPU
requirements, N1 and M1 sole-tenant node types are not recommended.
1Guest environment: For this OS, the guest environment can be manually
installed. To install the guest environment, see
Install the guest environment.
2gcloud CLI: For this OS, the gcloud CLI can be manually
installed. To install the gcloud CLI, see
Install the gcloud CLI.
3Suspend and resume: For this OS, some power policy configurations are
incompatible with suspend and resume. For more information, see
troubleshooting VM suspension.
4Suspend and resume: Suspend and resume for Windows 11 24h2 x64 & Windows 11
25h2 x64 isn't supported. This is a
known issue and a fix is in progress.
1Lifecycle stage: Google Cloud Support typically coincides with Extended Support End
Date. For information about the end of support (EOS) date for each of the operating systems in the
table below, see the
Microsoft Lifecycle Policy.
2Windows 11: vTPM and Secure Boot must be enabled. Due to CPU
requirements, N1 and M1 sole-tenant node types are not recommended.
1Guest environment: For this OS, the guest environment can be manually
installed. To install the guest environment, see
Install the guest environment.
2gcloud CLI: For this OS, the gcloud CLI can be manually
installed. To install the gcloud CLI, see
Install the gcloud CLI.
3Suspend and resume: For this OS, some power policy configurations are
incompatible with suspend and resume. For more information, see
troubleshooting VM suspension.
4Suspend and resume: Suspend and resume for Windows 11 24h2 x64 & Windows 11
25h2 x64 isn't supported. This is a
known issue and a fix is in progress.
Windows Server images are premium resources
that incur additional fees to use. Google Cloud builds and supports the Windows Server images
available for Compute Engine.
Windows Server images can't be used with the following:
Windows Server 2016 isn't supported with Compute Engine instances that have more than 255 vCPUs.
To bring your own Windows license (BYOL), you must either import an existing image, or you
must build a custom image.
Automatic updates
By default, this operating system is configured to "Auto download and schedule the install" for Microsoft updates.
To configure Windows Server automatic updates, see
Configure Automatic Updates.
Image configuration
Windows Server images are built with the latest updates, but have the following differences in
configuration from standard Windows Server images:
Account configuration
The Administrator account is disabled.
User passwords must be at least eight characters long.
The LocalAccountTokenFilterPolicy property is enabled to grant access to
administrative file shares.
Activation configuration
Windows Server images can't activate without a network connection to
kms.windows.googlecloud.com, and stop functioning if they don't authenticate
within 30 days. Make sure to [allow access in your VPC network](/compute/docs/instances/windows/creating-managing-windows-instances#kms-server).
A KMS client key is installed and the KMS client is set to activate by using the
Compute Engine KMS servers.
Emergency Management Services (EMS) redirection is enabled on the COM2 port. For more
information, see
bootcfg ems.
Network configuration
The Compute Engine metadata server is added to the hosts file, which is
typically in the %WinDir%\System32\drivers\etc directory.
The Windows firewall is open to allow communication with the Compute Engine metadata
server.
TCP KeepAliveTime is set to 5 minutes.
Web Proxy Auto Discovery (WPAD) is disabled.
The NetKVM adapter is set to use DHCP.
Remote Desktop (RDP) is enabled and the associated Windows firewall ports opened.
WinRM over HTTPS is configured using a self signed certificate and the associated Windows
firewall ports are open.
Windows images provided by Google have a hardcoded MTU. For more information about network and
interface MTU, see the maximum transmission unit overview.
Package system and Windows Update
Windows Server images update automatically from the Microsoft's Windows Update
service based on the default update schedule for Windows Server. To
receive updates, your Windows Server instances need internet access.
Google Cloud doesn't provide an internal Windows Update server.
To install packages for the guest
environment, Google Cloud repositories are enabled.
To manage Compute Engine component packages for Windows,
GooGet is installed, which you
can configure to update packages automatically.
The Google Cloud CLI is installed with its own Python 2.7
environment. Google Cloud CLI works with project service accounts, instance scopes, and works in
PowerShell and the standard command-line environment.
1Lifecycle stage: Google Cloud Support typically coincides with
Extended Support End Date. For
information about the end of support (EOS) date for each of the operating systems in the previous
table, see the
Microsoft Lifecycle Policy.
N1+GPU denotes support for NVIDIA T4, V100, or P4 GPUs running on a
general-purpose N1 machine family.
Limitations
You can't use Windows operating systems with the a2-megagpu-16g machine type.
When using A100 GPUs with a Windows operating system, choose a different A2 machine type.
You can't use Windows operating systems on g4-standard-384 machine types.
1Lifecycle stage: Google Cloud Support typically coincides with
Extended Support End Date. For
information about the end of support (EOS) date for each of the operating systems in the previous
table, see the
Microsoft Lifecycle Policy.
N1+GPU denotes support for NVIDIA T4, V100, or P4 GPUs running on a
general-purpose N1 machine family.
Limitations
You can't use Windows operating systems with the a2-megagpu-16g machine type.
When using A100 GPUs with a Windows operating system, choose a different A2 machine type.
You can't use Windows operating systems on g4-standard-384 machine types.
[[["Easy to understand","easyToUnderstand","thumb-up"],["Solved my problem","solvedMyProblem","thumb-up"],["Other","otherUp","thumb-up"]],[["Hard to understand","hardToUnderstand","thumb-down"],["Incorrect information or sample code","incorrectInformationOrSampleCode","thumb-down"],["Missing the information/samples I need","missingTheInformationSamplesINeed","thumb-down"],["Other","otherDown","thumb-down"]],["Last updated 2026-10-02 UTC."],[],[]]
