The storage-optimized machine family is suitable for workloads that are low in core usage and high in storage density. This machine family is ideal for high performance workloads that need fast access to data stored in local storage, such as data streaming, SQL and NoSQL databases, data search, data analytics, and data warehousing.
- The Z4M machine series offers compute instances with up to 192 vCPUs, up to 1,488 GB of memory, and up to 168,000 GiB of Titanium SSD capacity. Using technologies such as Titanium SSD, dual physical NICs, RDMA-enabled networking, and cluster management capabilities, these instances prioritize performance and workload-specific optimizations.
- The Z4D machine series offers compute instances with up to 3 TB of memory and 84,000 GiB of Titanium SSD capacity. The Z4D machine series includes machine types that provision 2 physical NICs per instance.
- The Z3 machine series offers compute instances with up to 1.5 TB of memory and up to 72,000 GiB of Titanium SSD capacity.
Z4D and Z3 offer machine types with different ratios of vCPU to Titanium SSD
capacity: standardlssd and highlssd.
| Machine series | Workloads |
|---|---|
| Z4M |
|
| Z4D |
|
| Z3 |
|
Z4M machine series
Z4M instances are powered by fifth generation Intel Emerald Rapids Xeon Scalable processors, DDR5 memory, and Titanium offload processors. Z4M machine types are optimized for the underlying NUMA architecture to deliver optimal, reliable, and consistent performance.
The Z4M machine series offers up to 168,000 GiB local storage using Titanium SSD.
Z4M instances use Titanium and Cloud RDMA to help enable higher levels of networking performance, isolation, and security.
- Z4M instances with 1 physical NIC support network bandwidths of up to 200 Gbps, depending on the number of vCPUs.
- Z4M instances with 2 physical NICs support network bandwidths of up to 400 Gbps.
Z4M offers the following features:
- Uses Titanium to offload networking and storage processing from the host CPU onto silicon devices deployed throughout the data center
- Delivers high performance block storage with Google Cloud Hyperdisk
- Offers the largest amount of Titanium SSD storage capacity for Compute Engine instances
- Supports Cloud RDMA
- Supports enhanced cluster management capabilities and can be used with AI Hypercomputer
- Supports Intel Advanced Matrix Extensions (AMX), which is an instruction set extension for CPUs to accelerate AI/ML inference and training workloads.
- Supports the following discount and consumption options:
- Resource-based committed use discounts (CUDs)
- Flexible CUDs
- Spot VMs (excluding bare metal machine types)
- Reservations
For details on pricing, see the VM pricing page. Disk usage and network usage is charged separately from machine type pricing. For more information, see Disk and image pricing and Network pricing. For Titanium SSD pricing, see Storage-optimized machine type family pricing.
Z4M limitations
The following restrictions apply:
- Z4M instances are available in select zones and regions.
- You can't use GPUs with Z4M instances.
- You can't suspend a Z4M instance.
- You can't create custom machine types for Z4M instances.
- Live migration is only supported on Z4M instances with 42,000 GiB or less of attached Titanium SSD. For larger instances or instances without live migration enabled, during maintenance the instance terminates with Titanium SSD data persistence.
- Live migration isn't supported on Z4M instances that enable Cloud RDMA. To enable Cloud RDMA, select Terminate for the host maintenance policy. During maintenance, these instances terminate with Titanium SSD data persistence.
- Z4M isn't supported on Windows OS images.
Z4M machine types
Z4M VMs are available as predefined configurations based on the highmem ratio
of 7.625 GB of memory per vCPU, or 7.75 GB of memory per vCPU for
larger machine types. The machine types offer from 16 to 192 vCPUs and up to
1,488 GB of memory. Z4M also offers a bare metal machine type with
192 vCPUs, 1,536 GB of memory, and 168,000 GiB of Titanium SSD.
| Machine types | vCPUs1 | Memory (GB) |
Titanium SSD | NUMA domains2 | Number of NICs | Default egress bandwidth (Gbps)3 |
Egress bandwidth (Gbps) outside a VPC network3 |
|---|---|---|---|---|---|---|---|
z4m-highmem-16 |
16 | 122 | (1 x 14,000 GiB) 14,000 GiB |
1 shared | 1 | Up to 32 | Up to 7 |
z4m-highmem-32 |
32 | 244 | (2 x 14,000 GiB) 28,000 GiB |
1 shared | 1 | Up to 67 | Up to 7 |
z4m-highmem-48 |
48 | 366 | (3 x 14,000 GiB) 42,000 GiB |
1 isolated | 1 | Up to 100 | Up to 25 |
z4m-highmem-192 |
192 | 1488 | (12 x 14,000 GiB) 168,000 GiB |
4 (whole host) | 2 | Up to 400 | Up to 50 |
z4m-highmem-192-metal |
192 | 1536 | (12 x 14,000 GiB) 168,000 GiB |
4 (whole host) | 2 | Up to 400 | Up to 50 |
1A vCPU is implemented as a single hardware thread on
the available CPU platform.
2 The number of vNUMA nodes exposed to the guest OS.
3 Maximum egress bandwidth can't exceed the number given. Actual
egress bandwidth depends on the destination IP address and other factors.
See Network bandwidth.
Every machine type in the Z4M machine series comes with locally attached Titanium SSD disks. The disks are added automatically when you create an instance. The capacity and performance for Titanium SSD disks for Z4M are listed in the following table:
| IOPS | Throughput (MiBps) |
||||||
|---|---|---|---|---|---|---|---|
| Machine type | # of attached Titanium disks |
Disk size (GiB) | Total size (GiB) | Read | Write | Read | Write |
z4m-highmem-16 |
1 | 14,000 | 14,000 | 1,300,000 | 1,024,000 | 6,300 | 4,800 |
z4m-highmem-32 |
2 | 14,000 | 28,000 | 2,600,000 | 2,048,000 | 12,600 | 9,600 |
z4m-highmem-48 |
3 | 14,000 | 42,000 | 3,900,000 | 3,072,000 | 18,900 | 14,400 |
z4m-highmem-192z4m-highmem-192-metal |
12 | 14,000 | 168,000 | 15,600,000 | 12,288,000 | 75,600 | 57,600 |
Supported disk types for Z4M
Z4M instances support only the NVMe disk interface and can use the following block storage types:
- Hyperdisk Balanced (
hyperdisk-balanced) - Hyperdisk Balanced High Availability (
hyperdisk-balanced-high-availability) - Hyperdisk Extreme (
hyperdisk-extreme) - Hyperdisk Throughput (
hyperdisk-throughput) - Hyperdisk ML (
hyperdisk-ml) - Titanium SSD
Disk and capacity limits
For details about the capacity limits, see Hyperdisk size and attachment limits and Persistent Disk maximum capacity.
The maximum number of disks that you can attach to Z4M instances are listed in the following tables:
| Machine type | Hyperdisk per VM | Hyperdisk Balanced | Hyperdisk Throughput | Hyperdisk Extreme | Hyperdisk ML |
|---|---|---|---|---|---|
z4m-highmem-16 |
32 | 32 | 32 | 0 | 32 |
z4m-highmem-32 |
32 | 32 | 32 | 0 | 32 |
z4m-highmem-48 |
32 | 32 | 32 | 8 | 32 |
z4m-highmem-192 |
128 | 128 | 64 | 8 | 64 |
z4m-highmem-192-metal |
32 | 32 | 32 | 8 | 32 |
For the performance limits of Hyperdisk, see Hyperdisk performance limits.
Network support for Z4M VMs
You can configure the following network interface types with Z4M instances:
- Z4M VM instances can have from 1 to 10 gVNIC network interfaces.
- Z4M bare metal instances can have from 1 to 2 IDPF network interfaces.
Z4M supports up to 400 Gbps network bandwidth between compute instances, and up to 50 Gbps egress bandwidth for Internet communication. Tier_1 networking isn't supported.
If you use Cloud RDMA, then you must specify a GVNIC or IDPF network interface before each IRDMA network interface when creating the instance. You can configure the following network interfaces (vNICs) when you create each instance:
Normal network traffic:
- For VM instances: You can have a maximum of 10 network interfaces. If you configure 1 IRDMA network interface, you can have from 1 to 9 GVNIC network interfaces. If you configure 2 IRDMA network interfaces, you can have from 1 to 8 GVNIC network interfaces.
- For bare metal instances: You can have 1 or 2 IDPF network interfaces per instance.
RDMA communication:
- You can configure up to 2 IRDMA network interfaces. You can't have more IRDMA network interfaces than GVNIC or IDPF network interfaces.
- IRDMA network interfaces are used only for Cloud RDMA communication.
- These network interfaces don't connect to the Internet.
Before migrating to Z4M or creating Z4M instances, make sure that the operating system image that you use is fully supported for Z4M. Fully supported images include support for 200 Gbps network bandwidth. If you are using Cloud RDMA, then the OS image must also support the IRDMA network interface type. If your Z4M instance is using an operating system that is not fully supported or has earlier versions of the network drivers, then your instance might not be able to achieve the maximum network bandwidth for Z4M instances.
The following table lists the network bandwidth limits for each Z4M machine type:
| Machine type | Maximum internal bandwidth (Gbps) | Maximum external bandwidth (Gbps) |
|---|---|---|
z4m-highmem-16 |
Up to 32 | Up to 7 |
z4m-highmem-32 |
Up to 67 | Up to 7 |
z4m-highmem-48 |
Up to 100 | Up to 25 |
z4m-highmem-192z4m-highmem-192-metal |
Up to 400 | Up to 50 |
Maintenance experience for Z4M instances
During the lifecycle of a Compute Engine instance, the host machine that your instance runs on undergoes multiple host events. A host event can include the regular maintenance of Compute Engine infrastructure, or in rare cases, a host error. Compute Engine also applies some non-disruptive lightweight upgrades for the hypervisor and network in the background.
The Z4M machine series supports on-demand maintenance and offers the following features related to host maintenance:
| Machine type | Typical scheduled maintenance event frequency | Maintenance behavior | Advanced notification | On-demand maintenance | Simulate maintenance |
|---|---|---|---|---|---|
z4m-highmem-16z4m-highmem-32z4m-highmem-48 |
Minimum of 30 days | Live migrate or terminate with Local SSD data persistence | 7 days | Yes | Yes |
z4m-highmem-192z4m-highmem-192-metal |
Minimum of 90 days | Terminates with Local SSD data persistence | 30 days | Yes | Yes |
The maintenance frequencies shown in the previous table are approximations, not guarantees. Compute Engine might occasionally perform maintenance more frequently.
Compute Engine preserves data on the local Titanium SSD disks for Z4M instances during maintenance events.
If a host event occurs, Compute Engine tries to recover any Titanium SSD disks attached to the instance. By default, Compute Engine spends up to 1 hour recovering the data. For Z4M VMs, Compute Engine spends up to 6 hours trying to recover the Titanium SSD data before reaching the timeout limit. For Z4M bare metal instances, Compute Engine spends up to 1 week (168 hours) trying to recover the data. This timeout limit is customizable. For more information about Local SSD and Titanium SSD recovery options, see Disk persistence following instance termination.
Z4D machine series
Z4D instances are powered by AMD EPYC Turin processors, DDR5 memory, and Titanium offload processors. Z4D machine types are optimized for the underlying NUMA architecture to deliver optimal, reliable, and consistent performance. Z4D also offers larger Titanium SSD disk sizes than Z3.
The Z4D machine series offers up to 84,000 GiB local storage using Titanium SSD.
Z4D offers the following features:
- Uses Titanium to offload networking and storage processing from the host CPU onto silicon devices deployed throughout the data center
- Delivers high performance block storage with Google Cloud Hyperdisk
- Offers a large amount of Titanium SSD storage capacity
- Supports Advanced Vector Extensions (AVX512), which is a built-in accelerator that offers built-in acceleration for vector-based processing, supporting 32 double-precision and 64 single-precision floating-point numbers, along with 8 64-bit and 16 32-bit integers.
- Z4D supports bare metal instances.
- Supports the following discount and consumption options:
- Resource-based committed use discounts (CUDs)
- Flexible CUDs
- Spot VMs (excluding bare metal machine types)
- Reservations
Z4D instances use Titanium to enable higher levels of networking performance, isolation, and security.
- Z4D instances with 1 physical NIC support network bandwidths of up to 200 Gbps, depending on the number of vCPUs.
- Z4D instances with 2 physical NICs support network bandwidths of up to 400 Gbps.
For details on pricing, see the VM pricing page. Disk usage and network usage is charged separately from machine type pricing. For more information, see Disk and image pricing and Network pricing. For Titanium SSD pricing, see Storage-optimized machine type family pricing.
Z4D Limitations
The following restrictions apply:
- Z4D instances are available in select zones and regions.
- You can't use GPUs with Z4D instances.
- Z4D doesn't support sole tenancy.
- You can't suspend a Z4D instance.
- You can't create custom machine types for Z4D instances.
- Live migration is only supported on Z4D instances with 42,000 GiB or less of attached Titanium SSD. For larger instances or instances without live migration enabled, during maintenance the instance terminates with Titanium SSD data persistence.
- Windows based operating system images aren't supported on Z4D instances.
- For regional availability of bare metal instances, see Bare metal instances.
Z4D machine types
The Z4D machine series supports the following predefined lssd machine
subtypes:
standardlssd: offers high performance search and data analysis for medium-sized data sets. This machine type has a vCPU to Titanium SSD capacity ratio of 1:219 and offers the highest Titanium SSD performance per vCPU.highlssd: offers high performance and storage intensive streaming and data analysis for large-sized data sets. This machine type has a vCPU to Titanium SSD capacity ratio of 1:438 and offers a larger Titanium SSD disk partition size thanstandardlssd.
Z4D standardlssd
| Machine types | vCPUs1 | Memory (GB) |
Titanium SSD | Number of NICs | Default egress bandwidth (Gbps)2 |
Egress bandwidth (Gbps) outside a VPC network2 |
|---|---|---|---|---|---|---|
z4d-highmem-16-standardlssd |
16 | 126 | (1 x 3,500 GiB) 3,500 GiB |
1 | Up to 20 | Up to 7 |
z4d-highmem-32-standardlssd |
32 | 252 | (2 x 3,500 GiB) 7,000 GiB |
1 | Up to 32 | Up to 7 |
z4d-highmem-48-standardlssd |
48 | 378 | (3 x 3,500 GiB) 10,500 GiB |
1 | Up to 50 | Up to 25 |
z4d-highmem-64-standardlssd |
64 | 504 | (4 x 3,500 GiB) 14,000 GiB |
1 | Up to 65 | Up to 25 |
z4d-highmem-96-standardlssd |
96 | 756 | (6 x 3,500 GiB) 21,000 GiB |
1 | Up to 100 | Up to 25 |
z4d-highmem-192-standardlssd |
192 | 1,512 | (12 x 3,500 GiB) 42,000 GiB |
1 | Up to 2003 | Up to 25 |
z4d-highmem-384-standardlssd |
384 | 3,024 | (24 x 3,500 GiB) 84,000 GiB |
2 | Up to 4003 | Up to 50 |
z4d-highmem-384-standardlssd-metal |
384 | 3,072 | (12 x 7,000 GiB) 84,000 GiB |
2 | Up to 400 | Up to 50 |
Z4D highlssd
| Machine types | vCPUs1 | Memory (GB) |
Titanium SSD | Number of NICs | Default egress bandwidth (Gbps)2 |
Egress bandwidth (Gbps) outside a VPC network2 |
|---|---|---|---|---|---|---|
z4d-highmem-8-highlssd |
8 | 63 | (1 x 3,500 GiB) 3,500 GiB |
1 | Up to 20 | Up to 7 |
z4d-highmem-16-highlssd |
16 | 126 | (2 x 3,500 GiB) 7,000 GiB |
1 | Up to 20 | Up to 7 |
z4d-highmem-32-highlssd |
32 | 252 | (4 x 3,500 GiB) 14,000 GiB |
1 | Up to 32 | Up to 7 |
z4d-highmem-48-highlssd |
48 | 378 | (6 x 3,500 GiB) 21,000 GiB |
1 | Up to 50 | Up to 25 |
z4d-highmem-64-highlssd |
64 | 504 | (8 x 3,500 GiB) 28,000 GiB |
1 | Up to 65 | Up to 25 |
z4d-highmem-96-highlssd |
96 | 756 | (12 x 3,500 GiB) 42,000 GiB |
1 | Up to 100 | Up to 25 |
z4d-highmem-192-highlssd |
192 | 1,512 | (24 x 3,500 GiB) 84,000 GiB |
1 | Up to 200 | Up to 25 |
1A CPU uses two threads per core, and a vCPU represents a single thread.
See CPU platforms.
2 Maximum egress bandwidth cannot exceed the number given. Actual
egress bandwidth depends on the destination IP address and other factors.
See Network bandwidth.
3 For instances with 192 or more vCPUs, you must configure at least two network
interfaces (vNICs), with each network interface attached to a different physical NIC, to
achieve the full network throughput. Compute Engine automatically maps the
vNIC to a physical NIC in a round-robin fashion.
Supported disk types for Z4D
Z4D instances support only the NVMe disk interface and can use the following block storage types:
- Hyperdisk Balanced (
hyperdisk-balanced) - Hyperdisk Balanced High Availability (
hyperdisk-balanced-high-availability) - Hyperdisk Extreme (
hyperdisk-extreme) - Hyperdisk Throughput (
hyperdisk-throughput) - Hyperdisk ML (
hyperdisk-ml) - Titanium SSD
Every machine type in the Z4D machine series comes with locally attached Titanium SSD disks. The disks are added automatically when you create an instance. The capacity and performance for Titanium SSD disks for Z4D are listed in the following table:
Z4D standardlssd
| IOPS | Throughput (MiBps) |
||||||
|---|---|---|---|---|---|---|---|
| Machine type | # of attached Titanium disks |
Disk size (GiB) | Total size (GiB) | Read | Write | Read | Write |
z4d-highmem-16-standardlssd |
1 | 3,500 | 3,500 | 650,000 | 512,000 | 3,150 | 2,400 |
z4d-highmem-32-standardlssd |
2 | 3,500 | 7,000 | 1,300,000 | 1,024,000 | 6,300 | 4,800 |
z4d-highmem-48-standardlssd |
3 | 3,500 | 10,500 | 1,950,000 | 1,536,000 | 9,450 | 7,200 |
z4d-highmem-64-standardlssd |
4 | 3,500 | 14,000 | 2,600,000 | 2,048,000 | 12,600 | 9,600 |
z4d-highmem-96-standardlssd |
6 | 3,500 | 21,000 | 3,900,000 | 3,072,000 | 18,900 | 14,400 |
z4d-highmem-192-standardlssd |
12 | 3,500 | 42,000 | 7,800,000 | 6,144,000 | 37,800 | 28,800 |
z4d-highmem-384-standardlssd |
24 | 3,500 | 84,000 | 15,600,000 | 12,288,000 | 75,600 | 57,600 |
z4d-highmem-384-standardlssd-metal |
12 | 7,000 | 84,000 | 15,600,000 | 12,288,000 | 75,600 | 57,600 |
Z4D highlssd
| IOPS | Throughput (MiBps) |
||||||
|---|---|---|---|---|---|---|---|
| Machine type | # of attached Titanium disks |
Disk size (GiB) | Total size (GiB) | Read | Write | Read | Write |
z4d-highmem-8-highlssd |
1 | 3,500 | 3,500 | 650,000 | 512,000 | 3,150 | 2,400 |
z4d-highmem-16-highlssd |
2 | 3,500 | 7,000 | 1,300,000 | 1,024,000 | 6,300 | 4,800 |
z4d-highmem-32-highlssd |
4 | 3,500 | 14,000 | 2,600,000 | 2,048,000 | 12,600 | 9,600 |
z4d-highmem-48-highlssd |
6 | 3,500 | 21,000 | 3,900,000 | 3,072,000 | 18,900 | 14,400 |
z4d-highmem-64-highlssd |
8 | 3,500 | 28,000 | 5,200,000 | 4,096,000 | 25,200 | 19,200 |
z4d-highmem-96-highlssd |
12 | 3,500 | 42,000 | 7,800,000 | 6,144,000 | 37,800 | 28,800 |
z4d-highmem-192-highlssd |
24 | 3,500 | 84,000 | 15,600,000 | 12,288,000 | 75,600 | 57,600 |
For the performance limits of Hyperdisk see Hyperdisk performance limits.
Disk and capacity limits
For details about the capacity limits, see Hyperdisk size and attachment limits.
The maximum number of disks that you can attach to Z4D instances are listed in the following tables:
Z4D standardlssd
| Machine type | Hyperdisk per VM | Hyperdisk Balanced | Hyperdisk Throughput | Hyperdisk Extreme | Hyperdisk ML |
|---|---|---|---|---|---|
z4d-highmem-16-standardlssd |
32 | 32 | 32 | 0 | 32 |
z4d-highmem-32-standardlssd |
32 | 32 | 32 | 0 | 32 |
z4d-highmem-48-standardlssd |
32 | 32 | 32 | 0 | 32 |
z4d-highmem-64-standardlssd |
64 | 64 | 64 | 8 | 64 |
z4d-highmem-96-standardlssd |
64 | 64 | 64 | 8 | 64 |
z4d-highmem-192-standardlssd |
64 | 64 | 64 | 8 | 64 |
z4d-highmem-384-standardlssd |
128 | 128 | 128 | 8 | 128 |
z4d-highmem-384-standardlssd-metal |
32 | 32 | 32 | 8 | 32 |
Z4D highlssd
| Machine type | Hyperdisk per VM | Hyperdisk Balanced | Hyperdisk Throughput | Hyperdisk Extreme | Hyperdisk ML |
|---|---|---|---|---|---|
z4d-highmem-8-highlssd |
16 | 16 | 16 | 0 | 16 |
z4d-highmem-16-highlssd |
32 | 32 | 32 | 0 | 32 |
z4d-highmem-32-highlssd |
32 | 32 | 32 | 0 | 32 |
z4d-highmem-48-highlssd |
32 | 32 | 32 | 0 | 32 |
z4d-highmem-64-highlssd |
64 | 64 | 64 | 8 | 64 |
z4d-highmem-96-highlssd |
64 | 64 | 64 | 8 | 64 |
z4d-highmem-192-highlssd |
64 | 64 | 64 | 8 | 64 |
Network support for Z4D VMs
The following network interface drivers are required:
- Z4D VM instances require GVNIC.
- Z4D bare metal instances require the Intel IDPF LAN PF device driver.
Z4D instances support up to 200 Gbps network bandwidth for standard networking. Z4D instances with two physical NICs support up to 400 Gbps network bandwidth for standard networking. Z4D instances don't use the per VM Tier_1 networking performance feature.
Z4D VM instances require gVNIC network interfaces.
Z4D supports up to 200 Gbps network bandwidth for standard networking. Z4D instances with two NICs support up to 400 Gbps network bandwidth for standard networking. Z4D instances don't use the per VM Tier_1 networking performance feature.
Before migrating to Z4D or creating Z4D VM instances, make sure that the operating system image that you use supports the IDPF network driver for bare metal instances or the gVNIC driver for VM instances. If your Z4D VM is using an operating system with an older version of gVNIC driver, this is still supported but the VM might experience suboptimal performance such as less network bandwidth or higher latency.
Google recommends using the latest gVNIC driver version to benefit from additional features and bug fixes.
Maintenance experience for Z4D instances
During the lifecycle of a Compute Engine instance, the host machine that your instance runs on undergoes multiple host events. A host event can include the regular maintenance of Compute Engine infrastructure, or in rare cases, a host error. Compute Engine also applies some non-disruptive lightweight upgrades for the hypervisor and network in the background.
The Z4D machine series supports on-demand maintenance and offers the following features related to host maintenance:
| Instance type | Typical scheduled maintenance event frequency | Maintenance behavior | Advanced notification | On-demand maintenance | Simulate maintenance |
|---|---|---|---|---|---|
| With 42,000 GiB or less attached Titanium SSD | Minimum of 30 days | Live migrate | 7 days | Yes | Yes |
| With more than 42,000 GiB attached Titanium SSD | Minimum of 30 days | Terminates with Local SSD data persistence | 7 days | Yes | Yes |
| Bare metal instances | Minimum of 30 days | Terminates with Local SSD data persistence | 7 days | Yes | Yes |
The maintenance frequencies shown in the previous table are approximations, not guarantees. Compute Engine might occasionally perform maintenance more frequently.
Compute Engine preserves data on the local Titanium SSD disks for Z4D instances during maintenance events.
If a host event occurs, Compute Engine tries to recover any Titanium SSD disks attached to the instance. By default, Compute Engine spends up to 1 hour recovering the data. For Z4D instances, Compute Engine spends up to 6 hours trying to recover the Titanium SSD data before reaching the timeout limit. This timeout limit is customizable. For more information about Local SSD and Titanium SSD recovery options, see Disk persistence following instance termination.
Z3 machine series
Z3 instances are powered by the fourth generation Intel Xeon Scalable processor (code-named Sapphire Rapids), DDR5 memory, and Titanium offload processors. Z3 machine types are optimized for the underlying NUMA architecture to deliver optimal, reliable, and consistent performance.
The Z3 machine series offers the following Local SSD storage capacities using Titanium SSD:
- Up to 36,000 GiB with VM instances
- 72,000 GiB with bare metal instances
Titanium SSD is custom-designed Local SSD based on Titanium I/O offload processing. It offers enhanced security, performance, and management compared to Local SSD.
Z3 offers the following features:
- Uses Titanium to offload networking and storage processing from the host CPU onto silicon devices deployed throughout the data center
- Delivers high performance block storage with Google Cloud Hyperdisk
- Offers a larger amount of Titanium SSD storage capacity than other Compute Engine machine series in different machine families
- Supports Intel Advanced Matrix Extensions (AMX), which is a built-in accelerator that significantly improves the performance of deep-learning training and inference on the CPU.
- Offers bare metal instances that provide access to several onboard, function-specific accelerators and offloads like Intel QAT, Intel DLB, Intel DSA, Intel TDX, and Intel IAA.
- Supports the following discount and consumption options:
- Resource-based committed use discounts (CUDs)
- Flexible CUDs
- Spot VMs (excluding bare metal machine types)
- Reservations
Z3 instances use Titanium to enable higher levels of networking performance, isolation, and security. The Z3 machine series supports a default network bandwidth of up to 100 Gbps and up to 200 Gbps with per VM Tier_1 networking performance.
For details on pricing, see the VM pricing page. Disk usage and network usage is charged separately from machine type pricing. For more information, see Disk and image pricing and Network pricing. For Titanium SSD pricing, see Storage-optimized machine type family pricing.
Z3 Limitations
The following restrictions apply:
- You can't use regional Persistent Disk with Z3 instances.
- Z3 instances are only available in select zones and regions. For regional availability of bare metal instances, see Bare metal instances.
- You can't use GPUs with Z3 instances.
- Z3 doesn't support sole tenancy.
- You can't suspend a Z3 instance.
- You can't create custom machine types for Z3 instances.
- Live migration is only supported for Z3 instances with 18,000 GiB or less of attached Titanium SSD.
- Windows based operating system images aren't supported on Z3 instances.
Z3 machine types
The Z3 machine series supports the following predefined lssd machine
subtypes:
standardlssd: offers high performance search and data analysis for medium-sized data sets. This machine type has a vCPU to Titanium SSD capacity ratio of less than 1:350 and offers the highest Titanium SSD performance per vCPU.highlssd: offers high performance and storage intensive streaming and data analysis for large-sized data sets. This machine type has a vCPU to Titanium SSD capacity ratio between 1:350 and 1:600 and offers a higher total Titanium SSD capacity thanstandardlssd.
To create a bare metal instance with Z3,
use the z3-highmem-192-highlssd-metal machine type.
Z3 standardlssd
| Machine types | vCPUs1 | Memory (GB) | Titanium SSD | Default egress bandwidth (Gbps)2 | Tier_1 egress bandwidth (Gbps)2 |
|---|---|---|---|---|---|
z3-highmem-14-standardlssd |
14 | 112 | (1 x 3000 GiB) 3,000 GiB | Up to 23 | N/A |
z3-highmem-22-standardlssd |
22 | 176 | (2 x 3000 GiB) 6,000 GiB | Up to 23 | N/A |
z3-highmem-44-standardlssd |
44 | 352 | (3 x 3000 GiB) 9,000 GiB | Up to 32 | Up to 50 |
z3-highmem-88-standardlssd |
88 | 704 | (6 x 3000 GiB) 18,000 GiB | Up to 62 | Up to 100 |
z3-highmem-176-standardlssd |
176 | 1,406 | (12 x 3000 GiB) 36,000 GiB | Up to 100 | Up to 200 |
1A CPU uses two threads per core, and a vCPU represents a single thread.
See CPU platforms.
2 Maximum egress bandwidth cannot exceed the number given. Actual
egress bandwidth depends on the destination IP address and other factors.
See Network bandwidth.
Z3 highlssd
| Machine types | vCPUs1 | Memory (GB) | Titanium SSD | Default egress bandwidth (Gbps)2 | Tier_1 egress bandwidth (Gbps)2 |
|---|---|---|---|---|---|
z3-highmem-8-highlssd |
8 | 64 | (1 x 3000 GiB) 3,000 GiB | Up to 23 | N/A |
z3-highmem-16-highlssd |
16 | 128 | (2 x 3000 GiB) 6,000 GiB | Up to 23 | N/A |
z3-highmem-22-highlssd |
22 | 176 | (3 x 3000 GiB) 9,000 GiB | Up to 23 | N/A |
z3-highmem-32-highlssd |
32 | 256 | (4 x 3000 GiB) 12,000 GiB | Up to 32 | N/A |
z3-highmem-44-highlssd |
44 | 352 | (6 x 3000 GiB) 18,000 GiB | Up to 32 | Up to 50 |
z3-highmem-88-highlssd |
88 | 704 | (12 x 3000 GiB) 36,000 GiB | Up to 62 | Up to 100 |
z3-highmem-192-highlssd-metal |
1923 | 1,536 | (12 x 6000 GiB) 72,000 GiB | Up to 100 | Up to 200 |
1A CPU uses two threads per core, and a vCPU represents a single thread.
See CPU platforms.
2 Maximum egress bandwidth cannot exceed the number given. Actual
egress bandwidth depends on the destination IP address and other factors.
See Network bandwidth.
3 For bare metal instances, the number of vCPUs is equivalent
to the number of hardware threads on the host server.
Supported disk types for Z3
Z3 VMs support only the NVMe disk interface and can use the following block storage types:
- Hyperdisk Balanced (
hyperdisk-balanced) - Hyperdisk Balanced High Availability (
hyperdisk-balanced-high-availability) - Hyperdisk Extreme (
hyperdisk-extreme) - Hyperdisk Throughput (
hyperdisk-throughput) - Balanced Persistent Disk (
pd-balanced) - SSD (performance) Persistent Disk (
pd-ssd) - Titanium SSD
Z3 bare metal instances can use the following block storage types:
- Hyperdisk Balanced (
hyperdisk-balanced) - Hyperdisk Extreme (
hyperdisk-extreme) - Titanium SSD
Every machine type in the Z3 machine series comes with locally attached Titanium SSD disks. The disks are added automatically when you create an instance. The capacity and performance for Titanium SSD disks for Z3 are listed in the following table:
Z3 standardlssd
| IOPS | Throughput (MiBps) |
||||||
|---|---|---|---|---|---|---|---|
| Machine type | # of attached Titanium disks |
Disk size (GiB) | Total size (GiB) | Read | Write | Read | Write |
z3-highmem-14-standardlssd |
1 | 3,000 | 3,000 | 750,000 | 500,000 | 3,000 | 2,500 |
z3-highmem-22-standardlssd |
2 | 3,000 | 6,000 | 1,500,000 | 1,000,000 | 6,000 | 5,000 |
z3-highmem-44-standardlssd |
3 | 3,000 | 9,000 | 2,250,000 | 1,500,000 | 9,000 | 7,500 |
z3-highmem-88-standardlssd |
6 | 3,000 | 18,000 | 4,500,000 | 3,000,000 | 18,000 | 15,000 |
z3-highmem-176-standardlssd |
12 | 3,000 | 36,000 | 9,000,000 | 6,000,000 | 36,000 | 30,000 |
Z3 highlssd
| IOPS | Throughput (MiBps) |
||||||
|---|---|---|---|---|---|---|---|
| Machine type | # of attached Titanium disks |
Disk size (GiB) | Total size (GiB) | Read | Write | Read | Write |
z3-highmem-8-highlssd |
1 | 3,000 | 3,000 | 750,000 | 500,000 | 3,000 | 2,500 |
z3-highmem-16-highlssd |
2 | 3,000 | 6,000 | 1,500,000 | 1,000,000 | 6,000 | 5,000 |
z3-highmem-22-highlssd |
3 | 3,000 | 9,000 | 2,250,000 | 1,500,000 | 9,000 | 7,500 |
z3-highmem-32-highlssd |
4 | 3,000 | 12,000 | 3,000,000 | 2,000,000 | 12,000 | 10,000 |
z3-highmem-44-highlssd |
6 | 3,000 | 18,000 | 4,500,000 | 3,000,000 | 18,000 | 15,000 |
z3-highmem-88-highlssd |
12 | 3,000 | 36,000 | 9,000,000 | 6,000,000 | 36,000 | 30,000 |
z3-highmem-192-highlssd-metal |
12 | 6,000 | 72,000 | 9,000,000 | 6,000,000 | 36,000 | 30,000 |
For the performance limits of Hyperdisk and Persistent Disk, see the following:
Disk and capacity limits
For details about the capacity limits, see Hyperdisk size and attachment limits and Persistent Disk maximum capacity.
Z3 storage limits are described in the following table:
Z3 standardlssd
| Maximum number of disks | |||||
|---|---|---|---|---|---|
| Machine type | Per VM1 | Hyperdisk volumes per VM |
Hyperdisk Balanced | Hyperdisk Throughput | Hyperdisk Extreme |
z3-highmem-14-standardlssd |
128 | 16 | 16 | 16 | 0 |
z3-highmem-22-standardlssd |
128 | 32 | 32 | 32 | 0 |
z3-highmem-44-standardlssd |
128 | 32 | 32 | 32 | 0 |
z3-highmem-88-standardlssd |
128 | 32 | 32 | 32 | 8 |
z3-highmem-176-standardlssd |
128 | 32 | 32 | 32 | 8 |
1 The maximum size per Hyperdisk volume is 64 TiB.
Z3 highlssd
| Maximum number of disks | |||||
|---|---|---|---|---|---|
| Machine type | Per VM1 | Hyperdisk volumes per VM |
Hyperdisk Balanced | Hyperdisk Throughput | Hyperdisk Extreme |
z3-highmem-8-highlssd |
128 | 16 | 16 | 16 | 0 |
z3-highmem-16-highlssd |
128 | 16 | 16 | 16 | 0 |
z3-highmem-22-highlssd |
128 | 32 | 32 | 32 | 0 |
z3-highmem-32-highlssd |
128 | 32 | 32 | 32 | 0 |
z3-highmem-44-highlssd |
128 | 32 | 32 | 32 | 0 |
z3-highmem-88-highlssd |
128 | 32 | 32 | 32 | 8 |
z3-highmem-192-highlssd-metal |
32 | 32 | 16 | 0 | 16 |
1 The maximum size per Hyperdisk volume is 64 TiB.
Network support for Z3 VMs
The following network interface drivers are required:
- Z3 VM instances require gVNIC.
- Z3 bare metal instances require the Intel IDPF LAN PF device driver.
Z3 supports up to 100 Gbps network bandwidth for standard networking and up to 200 Gbps with per VM Tier_1 networking performance for VM and bare metal instances.
Before migrating to Z3 or creating Z3 VMs or bare metal
instances, make sure that the
operating system image
that you use supports the IDPF network driver for bare metal instances or the
gVNIC driver for VM instances. To get the best possible performance on
Z3 VMs, choose an OS image that supports both
"Tier_1 Networking" and "200 Gbps network bandwidth". These images include an
updated gVNIC driver, even if the guest OS shows the gve driver version as
1.0.0. If your Z3 VM is
using an operating system with an older version of gVNIC driver, this is still
supported but the VM might experience suboptimal performance such as less
network bandwidth or higher latency.
If you use a custom OS image to create a Z3 VM, you can manually install the most recent gVNIC driver. The gVNIC driver version v1.4.2 or later is recommended for use with Z3 VMs. Google recommends using the latest gVNIC driver version to benefit from additional features and bug fixes.
Maintenance experience for Z3 instances
During the lifecycle of a Compute Engine instance, the host machine that your instance runs on undergoes multiple host events. A host event can include the regular maintenance of Compute Engine infrastructure, or in rare cases, a host error. Compute Engine also applies some non-disruptive lightweight upgrades for the hypervisor and network in the background.
The Z3 machine series supports on-demand maintenance and offers the following features related to host maintenance:
| Attached Titanium SSD (GiB) | Typical scheduled maintenance event frequency | Maintenance behavior | Advanced notification | On-demand maintenance | Simulate maintenance |
|---|---|---|---|---|---|
| 18,000 or less | Minimum of 30 days | Live migrate | 7 days | Yes | Yes |
| 36,000 | Minimum of 30 days | Terminates with Local SSD data persistence | 7 days | Yes | Yes |
| 72,000 (bare metal) | Minimum of 180 days | Terminates with Local SSD data persistence | 90 days | Yes | Yes |
The maintenance frequencies shown in the previous table are approximations, not guarantees. Compute Engine might occasionally perform maintenance more frequently.
Compute Engine preserves data on the local Titanium SSD disks for Z3 instances during maintenance events.
If a host event occurs, Compute Engine tries to recover any Titanium SSD disks attached to the instance. By default, Compute Engine spends up to 1 hour recovering the data. For Z3 instances, Compute Engine spends up to 6 hours trying to recover the Titanium SSD data before reaching the timeout limit. This timeout limit is customizable. For more information about Local SSD and Titanium SSD recovery options, see Disk persistence following instance termination.
What's next
- Creating and starting a virtual machine instance
- Learn about the different Storage options for your VM
- Move your workload to a new compute instance
- VM instance pricing