This document describes the features of the Compute Enginegeneral-purpose machine family, which has the best price-performance with themost flexible vCPU to memory ratios, and provides features that target moststandard and cloud-native workloads.

The general-purpose machine family has predefined andcustom machine types to align with your workload,depending on your requirements.

C4D is powered by the fifth generation AMD EPYC Turin processor andTitanium. These machine typeshave up to 384 vCPUs and 3,024 GB of DDR5 memory, a max-boostfrequency of 4.1 GHz, and up to 200 Gbps per VM Tier_1 networking performance.C4D also offers Local SSD (-lssd) machine types and bare metal (-metal)machine types.

C4A is powered by Google's Axion processor built on the Arm Neoverse V2 computecore. C4A provides machine types with up to 72 vCPUs,576 GB DDR5 memory, 6 TiB of localTitanium SSD, and up to 100 Gbps with per VM Tier_1 networking performance. C4A alsooffers Local SSD (-lssd) machine types.

C4 is powered by the sixth generation (code-named Granite Rapids) and fifthgeneration (code-named Emerald Rapids)Intel Xeon Scalable processors. C4 instances running on Granite Rapids offer asustained, all-core turbo frequencyof 3.9 GHz and a max turbo frequency of 4.2 GHz, 2.2 TB of DDR5 memory,18 TiB of local SSD, andsupports up to 200 Gbps ofper VM Tier_1 networking performance. C4 also offers Local SSD (-lssd) machine types and baremetal (-metal) machine types.

N4D is powered by the fifth generation AMD EPYC Turin processor andTitanium. Thesemachine types have up to 96 vCPUs and 768 GB of DDR5 memory, and amax-boost frequency of 4.1 GHz. N4D offers 50 Gbps of standard networkbandwidth.

N4A(Preview) is powered by Google'sAxion processor built on the Arm Neoverse N3 compute core. N4Aprovides machine types of up to 64 vCPUs and 512 GB of DDR5 memory.N4A is available in standard, high-mem, high-cpu, and custom machine types withextended memory, and up to 50 Gbps of standard networking.

N4 is powered by the fifth generation Intel Xeon Scalable processor(code-named Emerald Rapids). N4 offers a sustained, all-core turbo frequency of2.9 GHz, 640 GB of DDR5memory, and up to 50 Gbps of standard network bandwidth.

C3 is powered by fourth generation Intel Xeon Scalable processors and offers asustained, all-core turbo frequency of 3.0 GHz, 8 channels of DDR5 memory, andup to 200 Gbps per VM Tier_1 networking performance.

C3D is powered by fourth generation AMD EPYC Genoa processors and offers asustained, all-core turbo frequency of 3.3 GHz, 2,880 GB of DDR5 memory,and up to 200 Gbps per VM Tier_1 networking performance.

For bare metal machine types, choose the C4, C4D, or C3 machine series.

Allthird and fourth generationgeneral-purpose VMs supportTitanium.

E2, E2 shared-core, N2, N2D, Tau T2A, and Tau T2D are second generation machineseries in this family; N1 and its related shared-core machine types are thefirst generation machine series.

To learn how your selection affects the performance of Persistent Disk volumesattached to your VMs, seeConfigure your Persistent Disk and VMs.

C4D machine series

C4D VMs are powered by the fifth generation AMD EPYC Turin processor andTitanium.C4D delivers a 30% performance boost over C3D on the estimatedSPECrate®2017_int_base benchmark, which letsyou scale performance with fewer resources, thereby optimizing your costs.

C4D is designed to run workloads including web, app and game servers, AIinference, video streaming, and data centric applications likeanalytics, as well as relational and in-memory databases.

For databases, C4D delivers 55% more queries per second for MySQL and 35% higheroperations per second for Memorystore for Redis workloadscompared to C3D due to its higher core frequency (up to 4.1 GHz) andimproved Instructions Per Clock (IPC).

• Powered by the AMD EPYC Turin CPU and Titanium
• Supports up to 384 vCPUs and 3,024 GB of DDR5 memory
• Supports up to 12 TiB of local Titanium SSD disks
• Offers predefined machine types that range in size from 2 to 384 vCPUs
• Supports up to 3,024 GB of DDR5 memory for VM instances andup to 3,072 GB of memory for bare metal instances
• Supports consumption options like on-demand, Spot VMs,and future reservations
• Supports standard network configuration with up to 100 Gbps bandwidth
• Supports per VM Tier_1 networking performance with up to 200 Gbps bandwidth
• Supports only Hyperdisk volumes
• SupportsConfidential VM with AMD SEV
• Supportsresource-based and flexible committed use discounts (CUDs)
• Supportscompact and spread placement policies

C4D machine types

C4D VMs are available as predefined configurations instandard,highcpu,andhighmem sizes ranging from 2 vCPU to 384 vCPUs and up to 3,024 GBof memory.

To use Titanium SSD with C4D, create your instance using the-lssd variantof the C4D machine types. Selecting this machine type creates an instance of thespecified size with Titanium SSD partitions attached. You can't attachTitanium SSD volumes separately.

To create a bare metal instance with C4D, use one of the following machinetypes:

• c4d-standard-384-metal
• c4d-highcpu-384-metal
• c4d-highmem-384-metal

C4D doesn't support custom machine types.

Regional availability for C4D instances

For C4D VMs, you can view the available regions and zones in theAvailable regions and zones table,as follows:

• To view all the zones where you can create a C4D VM, in theSelect a machine series menu, selectC4D.
• You can also use theSelect a locationmenu to limit the results to a geographical area.

For regional availability of C4D bare metal instances, seeBare metal instances on Compute Engine.

Supported disk types for C4D

C4D VMs support only the NVMe diskinterface and can use the followingHyperdisk block storage:

• Hyperdisk Balanced (hyperdisk-balanced)
• Hyperdisk Extreme (hyperdisk-extreme)
• Local Titanium SSD(added automatically with-lssd machine types)

C4D doesn't support Persistent Disk.

Disk and capacity limits

You can attach a mixture of different Hyperdisk types toan instance, but the maximum total disk capacity (in TiB) across all disktypes can't exceed:

• For machine types with less than 32 vCPUs: 257 TiB for all Hyperdisk

• For machine types with 32 or more vCPUs: 512 TiB for all Hyperdisk

For details about the capacity limits, seeHyperdisk size and attachment limits.

C4D storage limits are described in the following table:

Network support for C4D instances

The following network interface drivers are required:

• C4D instances requiregVNIC network interfaces.
• C4D bare metal instances require theIntel IDPF LAN PF device driver.

C4D supports up to 100 Gbps network bandwidth for standardnetworking and up to 200 Gbps with per VM Tier_1 networking performance for VM andbare metal instances.

Before migrating to C4D or creating C4D VMs or bare metalinstances, make sure that theoperating system imagethat you use supports the IDPF network driver for bare metal instances or thegVNIC driver for VM instances. To get the best possible performance onC4D VMs, choose an OS image that supports both"Tier_1 Networking" and "200 Gbps network bandwidth". These images include anupdated gVNIC driver, even if the guest OS shows thegve driver version as1.0.0. If your C4D VM isusing an operating system with an older version of gVNIC driver, this is stillsupported but the VM might experience suboptimal performance such as lessnetwork bandwidth or higher latency.

If you use a custom OS image to create a C4D VM, you canmanually install the most recent gVNIC driver.The gVNIC driver version v1.4.2 or later is recommended for use with C4DVMs. Google recommends using the latest gVNIC driver version to benefit fromadditional features and bug fixes.

Maintenance experience for C4D instances

During the lifespan of a virtual machine (VM) instance,the host machine that your instance runs undergoes multiple host events.A host event can include the regular maintenance of Compute Engineinfrastructure, or in rare cases, a host error. Compute Engine alsoapplies some non-disruptive lightweight upgrades for the hypervisor and networkin the background.

The C4D machine series offers the following features related to hostmaintenance:

The maintenance frequencies shown in the previous table are approximations,not guarantees. Compute Engine might occasionally perform maintenancemore frequently.

C4A machine series

C4A VMs are powered by Google's first Arm Axion™ processor.C4A provides machine types with up to 72 vCPUs and 576 GB of DDR5memory, and 6 TiB of localTitanium SSD.C4A is available instandard,highmem, andhighcpu machine types, andalso offers-lssd variants for Titanium SSD. C4A uses Google Cloud's latestgeneration of storage options including Hyperdisk Balanced, Hyperdisk Extreme, and Titanium SSD.C4A offers up to 50 Gbps of standard network performance, and up to100 Gbps per VM Tier_1 networking performance for your instances.

C4A VMs are placed within a single node withUniform Memory Access (UMA)and also support sole tenant nodes to deliver consistent performance.

In summary, the C4A machine series has the following features:

• Is powered by the Google Axion CPU and Titanium.
• Supports up to 72 vCPUs and 576 GB of DDR5 memory.
• Supports up to 6 TiB of local Titanium SSD disks.
• Offers multiple predefined machine types.
• Supports standard network configuration with up to 50 Gbps bandwidth.
• Supports per VM Tier_1 networking performance with up to 100 Gbps bandwidth.
• Supports Hyperdisk only.
• Supports the following discount and consumption options:
  • Resource-based and flexible committed use discounts (CUDs)
  • Spot VMs
  • Reservations
• Supports theperformance monitoring unit (PMU).
• Doesn't supportcompact placement policies.
• Doesn't supportsuspendwith C4A instances that have attached Titanium SSD disks.

For information about migrating to Arm VMs, read theArm on Compute document.

C4A machine types

C4A VMs are available as predefined configurations insizes ranging from 1 vCPU to 72 vCPUs and up to 576 GB of memory.

• standard: 4 GB memory per vCPU
• highcpu: 2 GB memory per vCPU
• highmem: 8 GB memory per vCPU

To use Titanium SSD with C4A, create your VM using the-lssd variant ofthe C4A machine types. Selecting this machine type creates a VM of thespecified size with Titanium SSD partitions attached. You can't attachTitanium SSD volumes separately.

C4A doesn't support custom machine types.

Supported disk types for C4A

C4A VMs support only the NVMe disk interface and can use the followingHyperdisk block storage:

• Hyperdisk Balanced (hyperdisk-balanced)
• Hyperdisk Balanced High Availability (hyperdisk-balanced-high-availability)
• Hyperdisk Throughput (hyperdisk-throughput)
• Hyperdisk Extreme (hyperdisk-extreme)
• Local Titanium SSD(added automatically with-lssd machine types)

C4A doesn't support Persistent Disk.

Disk and capacity limits

You can attach a mixture of different Hyperdisk types toan instance, but the maximum total disk capacity (in TiB) across all disktypes can't exceed:

• For machine types with less than 32 vCPUs: 257 TiB for all Hyperdisk

• For machine types with 32 or more vCPUs: 512 TiB for all Hyperdisk

For details about the capacity limits, seeHyperdisk size and attachment limits.

Network support for C4A instances

C4A instances requiregVNIC network interfaces.C4A instances support up to 50 Gbps network bandwidth for standardnetworking and up to 100 Gbps network bandwidth per VM Tier_1 networking performance.

Before migrating to C4A or creating C4A instances,make sure that the operating system imagethat you use supports the gVNIC driver. To get the best possible performance onC4A instances, on theNetworking featurestab of the OS details table, choose an OS image that supports both"Tier_1 Networking" and "200 Gbps network bandwidth". These images include anupdated gVNIC driver, even if the guest OS shows thegve driver version as1.0.0. If your C4A instance is using an operating system with an olderversion of the gVNIC driver, this is still supported but the instance mightexperience suboptimal performance such as less network bandwidth or higherlatency.

If you use a custom OS image with the C4A machine series, you canmanuallyinstall the most recent gVNIC driver.The gVNIC driver version v1.4.2 or later is recommended for use with C4Ainstances. Google recommends using the latest gVNIC driver version to benefitfrom additional features and bug fixes.

Maintenance experience for C4A instances

During the lifespan of a virtual machine (VM) instance,the host machine that your instance runs undergoes multiple host events.A host event can include the regular maintenance of Compute Engineinfrastructure, or in rare cases, a host error. Compute Engine alsoapplies some non-disruptive lightweight upgrades for the hypervisor and networkin the background.

The C4A machine series offers the following features related to hostmaintenance:

The maintenance frequencies shown in the previous table are approximations,not guarantees. Compute Engine might occasionally perform maintenancemore frequently.

C4 machine series

C4 VMs are powered by 6th generation (code-named Granite Rapids) or 5thgeneration (code-named Emerald Rapids) Intel Xeon Scalable processors andTitanium. C4 Local SSD (-lssd) and bare metal(-metal) instances, as well asinstances with 144 or 288 vCPUs, use the 6th generation Intel Granite Rapidsprocessor. All other instances use the 5th generation Intel Emerald Rapidsprocessor.

The C4 machine series is designed to deliverprice-performance and enterprise-grade reliability along with a maintenanceexperience for your most demanding workloads. C4 instances are ideal for web andapp serving, game servers, databases andcaches, video streaming, data analytics, network appliances, and CPU-basedML inference.

C4 VMs are designed to achieve maximum performance from single-core turboboosting. For more consistent vCPU performance, disable vCPU boosting and limitthe vCPUs to the sustainable all-core turbo frequency. Youcan do this by settingturboMode=ALL_CORE_MAX in theAdvancedMachineFeaturessettings.

In summary, the C4 machine series:

• Is powered by the 6th generation Intel Granite Rapids or 5th generation IntelEmerald Rapids processor and Titanium IPU.
• Lets you switch between core-boosting performance and steady all-core turboperformance for your vCPUs.
• Supports up to 288 vCPUs and 2.2 TB of DDR5 memory.
• Supports up to 18 TiB of local Titanium SSD disks.
• Supports compact and spread placement policies.
• Offers multiple predefined machine types.
• Supports standard network configuration with up to 100 Gbps bandwidth.
• Supports per VM Tier_1 networking performance with up to 200 Gbps bandwidth.
• SupportsIntel Advanced Matrix Extensions (AMX),a built-in accelerator that significantly improves the performance ofdeep-learning training and inference on the CPU.
• Supports the following discount and consumption options:
  • Resource-based and flexible committed use discounts (CUDs)
  • Spot VMs
  • Reservations
• Supports theperformance monitoring unit (PMU).

C4 Limitations

• You can't dynamically add or remove a disk when using Windows Server 25.
• You can't dynamically add or remove multiple disks when using Windows Server25 or Windows 11.
• C4 VM shapes powered by Granite Rapidsmight experience lower networking performance on Windows 11 and Debian 11OS images.

C4 machine types

C4 VMs are available as predefined configurations insizes ranging from 2 vCPUs to 288vCPUs and up to 2,232 GBof memory.

• standard: 3.75 GB memory per vCPU
• highcpu: 2 GB memory per vCPU
• highmem: 7.75 GB memory per vCPU

To use Titanium SSD with C4, create your instance using the-lssd variant ofthe C4 machine types. Selecting this machine type creates an instance of thespecified size with Titanium SSD partitions attached. You can't attachTitanium SSD volumes separately.

To create a bare metal instance with C4,use one of the following machine types:

• c4-standard-288-metal
• c4-standard-288-lssd-metal
• c4-highmem-288-metal
• c4-highmem-288-lssd-metal

C4 doesn't support custom machine types.

Supported disk types for C4

C4 VMs support only the NVMe disk interface and can use the followingHyperdisk block storage:

C4 doesn't support Persistent Disk. When upgrading to a newer machine series, tomigrate your Persistent Disk resources to Hyperdisk, seeMove your workload from an existing VM to a new VM.

Disk and capacity limits

You can attach a mixture of different Hyperdisk types toan instance, but the maximum total disk capacity (in TiB) across all disktypes can't exceed:

• For machine types with less than 32 vCPUs: 257 TiB for all Hyperdisk

• For machine types with 32 or more vCPUs: 512 TiB for all Hyperdisk

For details about the capacity limits, seeHyperdisk size and attachment limits.

Network support for C4 VMs

The following network interface drivers are required:

• C4 instances requiregVNIC network interfaces.
• C4 bare metal instances require theIntel IDPF LAN PF device driver.

C4 supports up to 100 Gbps network bandwidth for standardnetworking and up to 200 Gbps with per VM Tier_1 networking performance for VM andbare metal instances.

Before migrating to C4 or creating C4 VMs or bare metalinstances, make sure that theoperating system imagethat you use supports the IDPF network driver for bare metal instances or thegVNIC driver for VM instances. To get the best possible performance onC4 VMs, choose an OS image that supports both"Tier_1 Networking" and "200 Gbps network bandwidth". These images include anupdated gVNIC driver, even if the guest OS shows thegve driver version as1.0.0. If your C4 VM isusing an operating system with an older version of gVNIC driver, this is stillsupported but the VM might experience suboptimal performance such as lessnetwork bandwidth or higher latency.

If you use a custom OS image to create a C4 VM, you canmanually install the most recent gVNIC driver.The gVNIC driver version v1.4.2 or later is recommended for use with C4VMs. Google recommends using the latest gVNIC driver version to benefit fromadditional features and bug fixes.

Maintenance experience for C4 instances

During thelifecycle of aCompute Engine instance, the host machine that your instance runs on undergoes multiplehost events.A host event can include the regular maintenance ofCompute Engine infrastructure, or in rare cases, a host error. Compute Engine alsoapplies some non-disruptive lightweight upgrades for the hypervisor and networkin the background.

The C4 machine series offers the following features related to hostmaintenance:

The maintenance frequencies shown in the previous table are approximations, not guarantees.Compute Engine might occasionally perform maintenance more frequently.

N4D machine series

N4D VMs are powered by the fifth generation AMD EPYC processors(code-name Turin) andTitanium. N4D VMs are engineered for flexibility,cost optimization, and enhanced price-performance through their efficientarchitecture. N4D supports next generation dynamic resource management, makingbetter use of resources on host machines.

In summary, the N4D machine series:

• Powered by the AMD EPYC Turin CPU and Titanium.
• Supports up to 96 vCPUs and 768 GB of DDR5 memory.
• Offers predefined machine types that range in size from 2 to 96 vCPUs.
• Supports custom machine types and extended memory.
• Supports consumption options like on-demand, Spot VMs,and future reservations.
• Supports standard network configuration with up to 50 Gbps bandwidth.
• Supports only Hyperdisk volumes.
• Supports resource-based and flexible committed use discounts(CUDs).
• Supportsspread placement policies.
• Doesn't support Local SSD or per per VM Tier_1 networking performance.

N4D machine types

N4D VMs are available as predefined configurations insizes ranging from 2 vCPUs to 96 vCPUs and up to 768 GB of memory.

• standard: 4 GB memory per vCPU
• highcpu: 2 GB memory per vCPU
• highmem: 8 GB memory per vCPU

Supported disk types for N4D

N4D VMs support only the NVMe diskinterface and can use the followingHyperdisk block storage:

• Hyperdisk Balanced (hyperdisk-balanced)
• Hyperdisk Balanced High Availability (hyperdisk-balanced-high-availability)
• Hyperdisk Throughput (hyperdisk-throughput)

N4D doesn't support Persistent Disk or Local SSD. ReadMove your workload from an existing VM to a new VMto migrate your Persistent Disk resources to a newer machine series.

Disk and capacity limits

The number of Hyperdisk volumes of all types that you canattach to a VM can't exceed the limits stated in theMax number ofHyperdisk volumes. For details about these limits, seeHyperdisk capacity.

For instances running Microsoft Windows and using the NVMe disk interface, thecombined number of both Hyperdisk and Persistent Diskattached volumes can't exceed a total of 16 disks.SeeKnown issues.Local SSD volumes are excluded from this issue.

N4D storage limits are described in the following table:

Network support for N4D VMs

N4D instances requiregVNIC network interfaces.N4D instances support up to 50 Gbps network bandwidth for standardnetworking and don't support per VM Tier_1 networking performance.

Before migrating to N4D or creating N4D VMinstances, make sure that theoperating system imagethat you use supports the gVNIC driver for VM instances.These images include anupdated gVNIC driver, even if the guest OS shows thegve driver version as1.0.0. If your N4D VM isusing an operating system with an older version of gVNIC driver, this is stillsupported but the VM might experience suboptimal performance such as lessnetwork bandwidth or higher latency.

If you use a custom OS image to create a N4D VM, you canmanually install the most recent gVNIC driver.The gVNIC driver version v1.4.2 or later is recommended for use with N4DVMs. Google recommends using the latest gVNIC driver version to benefit fromadditional features and bug fixes.

Maintenance experience for N4D instances

During thelifecycle of aCompute Engine instance, the host machine that your instance runs on undergoes multiplehost events.A host event can include the regular maintenance ofCompute Engine infrastructure, or in rare cases, a host error. Compute Engine alsoapplies some non-disruptive lightweight upgrades for the hypervisor and networkin the background.

The N4D machine series offers the following features related to hostmaintenance:

The maintenance frequencies shown in the previous table are approximations, not guarantees.Compute Engine might occasionally perform maintenance more frequently.

N4A machine series

This product or feature is subject to the "Pre-GA Offerings Terms" in the General Service Terms section of theService Specific Terms. Pre-GA products and features are available "as is" and might have limited support. For more information, see thelaunch stage descriptions.

N4A VMs are the second family of VMs powered by Google's latest custom-designedAxion processor, built on Arm Neoverse N3 compute core and powered byTitanium IPU. N4A VMs are placed within a single node withUniform Memory Access (UMA).They are engineered to be our most efficient and flexible Arm VMs,delivering exceptional price-performance for a wide range of general-purposeand scale-out workloads. N4A uses next generationdynamic resource management,which makes better use of resources on host machines.

Ideal use cases include web and application servers,microservices, containerized applications using Google Kubernetes Engine (GKE), open-sourcedatabases, and development and testing environments.

In summary, the N4A machine series:

• Is powered by the Google Axion Arm processor and Titanium IPU.
• Supports up to 64 vCPUs and 512 GB of DDR5 memory.
• Offers multiple predefined machine types andcustom machine types with extended custom memory upto 512 GB.
• Supports standard network configuration with up to 50 Gbps of bandwidth.
• Supports Hyperdisk only.
• Supports the following discount and consumption options:
  • Resource-based and flexible committed use discounts (CUDs)
  • Spot VMs
  • Reservations
• Doesn't support Local SSD or per VM Tier_1 networking performance.
• Confidential VM is not supported by this CPU.
• 32-bit mode EL0 (guest userspace) is not supported due to a hardwarelimitation.

N4A machine types

N4A VMs are available as predefined configurations insizes ranging from 1 vCPUs to 64 vCPUs and up to 512 GB of memory.

• standard: 4 GB memory per vCPU
• highcpu: 2 GB memory per vCPU
• highmem: 8 GB memory per vCPU

For information about custom machine types, seeCustom machine types.

Supported disk types for N4A

N4A VMs support only the NVMe disk interface and can use the followingHyperdisk block storage:

• Hyperdisk Balanced (hyperdisk-balanced)
• Hyperdisk Balanced High Availability (hyperdisk-balanced-high-availability)
• Hyperdisk Throughput (hyperdisk-throughput)

N4A doesn't support Persistent Disk or Local SSD. ReadMove your workload from an existing VM to a new VMto migrate your Persistent Disk resources to a newer machine series.

Disk and capacity limits

The number of Hyperdisk volumes of all types that you canattach to a VM can't exceed the limits stated in theMax number ofHyperdisk volumes. For details about these limits, seeHyperdisk capacity.

The combined total number of Hyperdisk Balanced volumes attached to a single VM depends onthe number of vCPUs the VM has, and can't exceed these limits:

N4A storage limits are described in the following table:

Network support for N4A VMs

N4A instances requiregVNIC network interfaces.N4A instances support up to 50 Gbps network bandwidth for standardnetworking and don't support per VM Tier_1 networking performance.

Before migrating to N4A or creating N4A VMinstances, make sure that theoperating system imagethat you use supports the gVNIC driver for VM instances.These images include anupdated gVNIC driver, even if the guest OS shows thegve driver version as1.0.0. If your N4A VM isusing an operating system with an older version of gVNIC driver, this is stillsupported but the VM might experience suboptimal performance such as lessnetwork bandwidth or higher latency.

If you use a custom OS image to create a N4A VM, you canmanually install the most recent gVNIC driver.The gVNIC driver version v1.4.2 or later is recommended for use with N4AVMs. Google recommends using the latest gVNIC driver version to benefit fromadditional features and bug fixes.

Maintenance experience for N4A instances

During thelifecycle of aCompute Engine instance, the host machine that your instance runs on undergoes multiplehost events.A host event can include the regular maintenance ofCompute Engine infrastructure, or in rare cases, a host error. Compute Engine alsoapplies some non-disruptive lightweight upgrades for the hypervisor and networkin the background.

The N4A machine series offers the following features related to hostmaintenance:

N4 machine series

N4 VMs are powered by the 5th generation Intel Xeon Scalable processors(code-named Emerald Rapids) andTitanium. N4 machine types are builtfrom the ground up for flexibility and cost optimization through an efficientarchitecture of streamlined features, shapes, and next generationdynamic resource management, whichmakes better use of resources on host machines. N4 offers flexible optionslike custom machine types that lets you use choose varied combinations ofcompute and memory to optimize costs and reduce resource waste. N4 is suitedfor a variety of general-purpose workloads that don't require peak processingpower at all times.

In summary, the N4 machine series:

• Is powered by 5th generation Intel Emerald Rapids processor and titaniumprocessors.
• Supports up to 80 vCPUs and 640 GB of DDR5 memory.
• Offers multiple predefined machine types andcustom machine types and extendedcustom memory up to 640 GB.
• Supports standard network configuration with up to 50 Gbps bandwidth
• Supports Intel Advanced Matrix Extensions (AMX), a built-in accelerator thatsignificantly improves the performance of deep-learning training and inference on the CPU.
• Supports the following discount and consumption options:
  • Resource-based and flexible committed use discounts (CUDs)
  • Spot VMs
  • Reservations
• Doesn't support Local SSD or per VM Tier_1 networking performance.

N4 machine types

N4 VMs are available as predefined configurations insizes ranging from 2 vCPUs to 80 vCPUs and up to 640 GB of memory.

• standard: 4 GB memory per vCPU
• highcpu: 2 GB memory per vCPU
• highmem: 8 GB memory per vCPU

Supported disk types for N4

N4 VMs supports only the NVMe disk interface and can use the followingHyperdisk block storage:

• Hyperdisk Balanced (hyperdisk-balanced)
• Hyperdisk Balanced High Availability (hyperdisk-balanced-high-availability)
• Hyperdisk Throughput (hyperdisk-throughput)

N4 doesn't support Persistent Disk or Local SSD. ReadMove your workload from an existing VM to a new VMto migrate your Persistent Disk resources to a newer machine series.

Disk and capacity limits

The number of Hyperdisk volumes of all types that you canattach to a VM can't exceed the limits stated in theMax number ofHyperdisk volumes. For details about these limits, seeHyperdisk capacity.

N4 storage limits are described in the following table:

Network support for N4 VMs

N4 instances requiregVNIC network interfaces.N4 instances support up to 50 Gbps network bandwidth for standardnetworking and don't support per VM Tier_1 networking performance.

Before migrating to N4 or creating N4 VMinstances, make sure that theoperating system imagethat you use supports the gVNIC driver for VM instances.These images include anupdated gVNIC driver, even if the guest OS shows thegve driver version as1.0.0. If your N4 VM isusing an operating system with an older version of gVNIC driver, this is stillsupported but the VM might experience suboptimal performance such as lessnetwork bandwidth or higher latency.

If you use a custom OS image to create a N4 VM, you canmanually install the most recent gVNIC driver.The gVNIC driver version v1.4.2 or later is recommended for use with N4VMs. Google recommends using the latest gVNIC driver version to benefit fromadditional features and bug fixes.

Maintenance experience for N4 instances

During thelifecycle of aCompute Engine instance, the host machine that your instance runs on undergoes multiplehost events.A host event can include the regular maintenance ofCompute Engine infrastructure, or in rare cases, a host error. Compute Engine alsoapplies some non-disruptive lightweight upgrades for the hypervisor and networkin the background.

The N4 machine series offers the following features related to hostmaintenance:

The maintenance frequencies shown in the previous table are approximations, not guarantees.Compute Engine might occasionally perform maintenance more frequently.

C3D machine series

C3D VMs are powered by the 4th generation AMD EPYC™ (Genoa) processor witha maximum frequency of 3.7 Ghz. C3D machine types are optimized for theunderlying hardware architecture to deliver optimal, reliable, and consistentperformance.

C3D usesTitanium, which enables higher levels of networking performance, isolation and security. The C3D machine series supports Tier_1 networking bandwidth of up to 100 Gbps and up to 200 Gbps.

In summary, the C3D machine series:

• Is powered by 4th generation AMD EPYC™ processor and Titanium.
• Supports up to 360 vCPUs and 2,880 GB of DDR5 memory.
• Supports standard network configuration with up to 100 Gbps bandwidthand Tier_1 networking with up to 200 Gbps bandwidth.
• Supports the following discount and consumption options:
  • Resource-based and flexible committed use discounts (CUDs)
  • Spot VMs
  • Reservations
• SupportsConfidential VMwith AMD SEV.

C3D machine types

C3D VMs are available instandard,highcpu,highmem, andlssdconfigurations in sizes ranging from 4 to 360 vCPUs and up to 2,880 GB ofmemory. Thehighcpu configuration offers the lowest price per performance forcompute-bound workloads that don't require large amounts of memory.

C3D doesn't support custom machine types.

Supported disk types for C3D

C3D VMs support only the NVMe disk interface and can use the following blockstorage types:

• Balanced Persistent Disk (pd-balanced)
• SSD (performance) Persistent Disk (pd-ssd)
• Hyperdisk Balanced (hyperdisk-balanced)
• Hyperdisk Balanced High Availability (hyperdisk-balanced-high-availability)
• Hyperdisk ML (hyperdisk-ml)
• Hyperdisk Extreme (hyperdisk-extreme)
• Hyperdisk Throughput (hyperdisk-throughput)
• Local SSD (only available with-lssd machine types)

To use Local SSD with C3D, create your VM using the-lssd variant of theC3D machine types. Selecting this machine type creates a VM of the specifiedsize with Local SSD partitions attached. You must use a machine type that endsin-lssd to use Local SSD with your C3D VM; you can't attach Local SSD volumesseparately.

Disk and capacity limits

For instances running Microsoft Windows and using the NVMe disk interface, thecombined number of both Hyperdisk and Persistent Diskattached volumes can't exceed a total of 16 disks.SeeKnown issues.Local SSD volumes are excluded from this issue.

C3D storage limits are described in the following table:

Network support for C3D VMs

C3D instances requiregVNIC network interfaces.C3D supports up to 100 Gbps network bandwidth for standardnetworking and up to 200 Gbps with per VM Tier_1 networking performance.

Before migrating to C3D or creating C3D instances,make sure that the operating system imagethat you use supports the gVNIC driver. To get the best possible performance onC3D instances, on theNetworking featurestab of the OS details table, choose an OS image that supports both"Tier_1 Networking" and "200 Gbps network bandwidth". These images include anupdated gVNIC driver, even if the guest OS shows thegve driver version as1.0.0. If your C3D instance is using an operating system with an olderversion of the gVNIC driver, this is still supported but the instance mightexperience suboptimal performance such as less network bandwidth or higherlatency.

If you use a custom OS image with the C3D machine series, you canmanuallyinstall the most recent gVNIC driver.The gVNIC driver version v1.4.2 or later is recommended for use with C3Dinstances. Google recommends using the latest gVNIC driver version to benefitfrom additional features and bug fixes.

Maintenance experience for C3D instances

During thelifecycle of aCompute Engine instance, the host machine that your instance runs on undergoes multiplehost events.A host event can include the regular maintenance ofCompute Engine infrastructure, or in rare cases, a host error. Compute Engine alsoapplies some non-disruptive lightweight upgrades for the hypervisor and networkin the background.

The C3D machine series offers the following features related to hostmaintenance:

The maintenance frequencies shown in the previous table are approximations, not guarantees.Compute Engine might occasionally perform maintenance more frequently.

C3 machine series

C3 VMs are powered by the 4th generation Intel Xeon Scalable processors(code-named Sapphire Rapids), DDR5 memory, andTitanium.C3 machine types are optimized for the underlying NUMA architecture to deliveroptimal, reliable, and consistent performance.

The new C3 machine series is a major leap in our purpose-built infrastructureofferings:

• Leveraging Titanium processors to offload networking from the CPUs
• Delivering high performance block-storage withGoogle Cloud Hyperdisk
• Speeding up ML training and inference withIntel AMX

C3 uses Titanium to enable higher levels of networking performance, isolationand security. The C3 machine series supports a default network bandwidth of upto 100 Gbps and up to 200 Gbps withper VM Tier_1 networking performance.Titanium has been designed from the ground up to enable updates that don'timpact running workloads.

The C3 machine series provides some of the largest general-purpose machinetypes, letting you create VM instances with up to 176 vCPUs and 1.4 TB ofmemory.

C3 has bare metal machine types,which allow you to access all the raw compute resources of the server. You cancreate bare metal instances with 192 vCPUs and up to 1,536 GB of memory.Bare metal instances also provide access to several onboard, function-specificaccelerators and offloads:

• Intel-QAT
• Intel-DLB
• Intel DSA
• Intel IAA

If your organization uses a Shielded VM policy, then you must createa custom org policy that excludes bare metal shapes before you can create abare metal instance.

In summary, the C3 machine series:

• Is powered by Intel 4th Generation Xeon processors and Titanium.
• Supports up to 176 vCPUs and 1.4 TB of DDR5 memory for VMs.
• Supports up to 192 vCPUs and 1,536 GB of memory for bare metalinstances.
• Supports standard network configuration with up to 100 Gbps bandwidthand Tier_1 networking with up to 200 Gbps bandwidth.
• Supports Intel Advanced Matrix Extensions (AMX), a built-in acceleratorthat significantly improves the performance of deep-learning training andinference on the CPU.
• Supports the following discount and consumption options:
  • Resource-based and flexible committed use discounts (CUDs)
  • Spot VMs
  • Reservations
• SupportsConfidential VMwith Intel TDX.
• Doesn't offersustained use discounts (SUDs).
• C3 bare metal instances don't support the following:
  • Google Kubernetes Engine
  • Shielded VM
  • Nested virtualization

Caution: When you purchase resource-based commitments for C3 and C3D resources, the machine family that is specified by the commitment type changes depending on the interface:

• In the gcloud CLI and REST, the commitment type values useCompute-optimized as the machine family, even though C3 and C3D are part of the general-purpose machine family.
• In the Google Cloud console, the commitment type values use the correct machine series:General-Purpose.

C3 machine types

C3 VMs are available in predefined machine types with sizes ranging from 4 to176 vCPUs and up to 1,408 GB of memory.

To use Local SSD with C3, create your VM using the-lssd variant of theC3 machine types. Selecting this machine type creates a VM of the specifiedsize with Local SSD partitions attached. You must use ac3-standard-*-lssdmachine type to use Local SSD with your VM; you can't attach Local SSD volumesseparately.

To create a bare metal instance with C3, use one of the following machine types:

• c3-standard-192-metal
• c3-highcpu-192-metal
• c3-highmem-192-metal

C3 doesn't support custom machine types.

C3 regional availability for bare metal instances

For C3 VMs, you can view the available regions and zones in theAvailable regions and zones table.

• SelectC3 in theSelect a machine type drop-down menu to see all thezones where you can create a C3 VM.
• You can also use theSelect a locationdrop-down menu to limit the results to a geographical area.

C3 bare metal instances are available in the following regions and zones:

Supported disk types for C3

C3 VMs support only the NVMe disk interface and can use the following blockstorage types:

A set amount of Local SSD disks are added to the C3 VM when you use the-lssdmachine type. This is the only way to include Local SSD storage with a C3 VM.You can't use Local SSD disks with bare metal instances.

Disk and capacity limits

For instances running Microsoft Windows and using the NVMe disk interface, thecombined number of both Hyperdisk and Persistent Diskattached volumes can't exceed a total of 16 disks.SeeKnown issues.Local SSD volumes are excluded from this issue.

C3 storage limits are described in the following table:

Network support for C3 VMs

The following network interface drivers are required:

• C3 instances requiregVNIC network interfaces.
• C3 bare metal instances require theIntel IDPF LAN PF device driver.

C3 supports up to 100 Gbps network bandwidth for standardnetworking and up to 200 Gbps with per VM Tier_1 networking performance for VM andbare metal instances.

Before migrating to C3 or creating C3 VMs or bare metalinstances, make sure that theoperating system imagethat you use supports the IDPF network driver for bare metal instances or thegVNIC driver for VM instances. To get the best possible performance onC3 VMs, choose an OS image that supports both"Tier_1 Networking" and "200 Gbps network bandwidth". These images include anupdated gVNIC driver, even if the guest OS shows thegve driver version as1.0.0. If your C3 VM isusing an operating system with an older version of gVNIC driver, this is stillsupported but the VM might experience suboptimal performance such as lessnetwork bandwidth or higher latency.

If you use a custom OS image to create a C3 VM, you canmanually install the most recent gVNIC driver.The gVNIC driver version v1.4.2 or later is recommended for use with C3VMs. Google recommends using the latest gVNIC driver version to benefit fromadditional features and bug fixes.

Maintenance experience for C3 instances

During thelifecycle of aCompute Engine instance, the host machine that your instance runs on undergoes multiplehost events.A host event can include the regular maintenance ofCompute Engine infrastructure, or in rare cases, a host error. Compute Engine alsoapplies some non-disruptive lightweight upgrades for the hypervisor and networkin the background.

The C3 machine series offers the following features related to hostmaintenance:

The maintenance frequencies shown in the previous table are approximations, not guarantees.Compute Engine might occasionally perform maintenance more frequently.

N2D machine series

The N2D machine series can run on either AMD EPYC Milan or AMD EPYCRome processors. The third generationAMD EPYC Milan processor is available only in specificregions and zones.To useAMD Milan as yourminimum CPU platform,request it when you create your VM instance.

The N2D series provides some of the largest general-purpose machine types withup to 224 vCPUs and 896 GB of memory and vCPU to memory ratios of 1:1, 1:4, and1:8. The AMD EPYC Rome processors in this series run with a base frequency of2.25 GHz, an effective frequency of 2.7 GHz, and a max boost frequency of3.3 GHz.

In summary, the N2D series:

• Support up to 224 vCPUs and 896 GB of memory.
• Support 50 Gbps and 100 Gbpshigh-bandwidth network configurations.
• Available in predefined andcustom VMs.
• Offer higher memory-to-core ratios for VMs created with the extended memoryfeature. Using the extended memory feature helps you avoid per-CPU softwarelicensing costs while providing access to more than 8 GB of memory per vCPU.
• Powered by third generation AMD EPYC Milan and second generation AMD EPYCRome processors.
• Supports the following discount and consumption options:
  • Resource-based and flexible committed use discounts (CUDs)
  • Sustained use discounts (SUDs)
  • Spot VMs
  • Reservations
• Doesn't support GPUs or nested virtualization.
• SupportsConfidential VMwith AMD SEV and AMD SEV-SNP.

N2D VMs don't support GPUs or nested virtualization.

N2D machine types

The following table lists the features of the N2D machine series. For somemachine types, certain features are not applicable (N/A).

The amount of memory configured per vCPU differs depending on the machine type:

• standard: 4 GB of system memory per vCPU
• highmem: 8 GB of system memory per vCPU
• highcpu: 1 GB of system memory per vCPU

For details on the pricing information, see the following:

• For machine type pricing, seeVM pricing page.
• Disk usage and network usage is charged separately from machine type pricing.For details, seeDisk and imagepricing andNetwork pricing.
• For per VM Tier_1 network performance billing rates, seeTier_1 higher bandwidth network pricing.

Supported disk types for N2D

N2D VMs can use the following block storage types:

• Zonal and regional standard Persistent Disk (pd-standard)
• Zonal and regional balanced Persistent Disk (pd-balanced)
• Zonal and regional SSD Persistent Disk (pd-ssd)
• Hyperdisk Throughput (hyperdisk-throughput)
• Local SSD

N2 machine series

The N2 machine series has flexible sizing between 2 to 128 vCPUs and 0.5 to8 GB of memory per vCPU. Machine types in this series run on thefollowing processors:

• Ice Lake—offered in specificregions and zones. It isthe default processor for larger machine types.

• Cascade Lake—the default for machine types up to 80 vCPUs. If you wantto create VMs withIce Lake, you must set it as theminimum CPU platform.

You can find more details about these two processors on theCPU platforms page.

Workloads that can take advantage of the higher clock frequency are a goodchoice for this series. These workloads can get higher per-threadperformance while benefiting from all the flexibility that the general-purposemachine family offers.

In summary, the N2 machine series:

• Supports up to 128 vCPUs and 864 GB of memory.
• Supports 50 Gbps, 75 Gbps, and 100 Gbpshigh-bandwidth network configurations.
• Is available in predefined andcustom VMs.
• Has higher memory-to-core ratios for VMs created with the extended memoryfeature. Using the extended memory feature helps control per-CPU softwarelicensing costs while providing access to more than 8 GB of memory pervCPU.
• Supports the following discount and consumption options:
  • Resource-based and flexible committed use discounts (CUDs)
  • Sustained use discounts (SUDs)
  • Spot VMs
  • Reservations

N2 machine types

The amount of memory configured per vCPU differs depending on the machine type:

• standard: 4 GB of system memory per vCPU
• highmem: 8 GB of system memory per vCPU
• highcpu: 1 GB of system memory per vCPU