To create a Confidential VM instance, you need a virtual machine that has the
following properties:

- [A supported Compute Engine machine type, CPU, and zone](https://docs.cloud.google.com/confidential-computing/confidential-vm/docs/supported-configurations#machine-types)

- [A supported operating system](https://docs.cloud.google.com/confidential-computing/confidential-vm/docs/supported-configurations#operating-systems)

You can configure your own Confidential VM instance manually, or accept the
suggested settings when you enable the Confidential VM service in the
Google Cloud console.

## Limitations and known issues

The following limitations apply depending on how you've configured your
Confidential VM instance.

### All Confidential VM instances

- You must create a new VM instance to enable Confidential VM. Existing
  instances can't be converted to Confidential VM instances.

- You can't attach [TPUs](https://docs.cloud.google.com/tpu/docs/intro-to-tpu) to Confidential VM instances.

- Confidential VM instances require an NVMe interface for disks. SCSI isn't
  supported.

- Only new disks can be formatted to XFS on Linux kernel versions earlier
  than 5.10. To format existing disks to XFS, you need kernel version 5.10
  or later.

- You can't attach more than 40 disks to a Confidential VM instance.
  You can request an exception through a
  [support channel](https://docs.cloud.google.com/confidential-computing/confidential-vm/docs/get-support), although instances with
  more than 40 disks might fail silently.

- Boot time is proportional to the amount of memory assigned to an instance.
  You might notice longer boot times for Confidential VM instances with large
  amounts of memory.

- Establishing SSH connections takes longer on Confidential VM instances than
  non-Confidential VM instances.

- [Live migration](https://docs.cloud.google.com/confidential-computing/confidential-vm/docs/troubleshoot-live-migration) is
  supported on N2D and C3D machine types running AMD SEV.

- Confidential VM instances might experience lower network bandwidth and higher
  latency compared to non-Confidential VM instances.

- Confidential VM instances with Hyperdisk Throughput have a restricted maximum data
  transfer size (MDTS), which reduces performance. To minimize performance
  degradation, avoid read-ahead requests exceeding the MDTS limit by setting
  `read_ahead_kb` to `124` KiB when using buffered I/O.

### AMD SEV

- Debian 12 doesn't have attestation support for AMD SEV due to the missing
  `/dev/sev-guest` package.

- AMD SEV on C2D and N2D machine types has a maximum
  [vNIC queue count](https://docs.cloud.google.com/compute/docs/network-bandwidth#rx-tx) of `8`.

<br />

- AMD SEV on the C4D and C3D machine type has the following limitations:

  - Confidential VM instances that use C4D and C3D machine types might
    experience lower network bandwidth than equivalent non-confidential VM
    instances, even with
    [per VM Tier_1 networking performance](https://docs.cloud.google.com/compute/docs/networking/configure-vm-with-high-bandwidth-configuration)
    enabled.

  - Bare metal instances aren't supported.

  - The `rhel-8-4-sap-ha` image tagged `SEV_CAPABLE` doesn't work with AMD
    SEV on C4D and C3D machines with more than 8 vCPUs. This image is
    missing a required
    [patch](https://kernel.googlesource.com/pub/scm/linux/kernel/git/torvalds/linux.git/+/e998879d4fb7991856916972168cf27c0d86ed12%5E%21/?pli=1)
    that increases the size of the SWIOTLB buffer for high networking
    queues.

  - Confidential VM instances with AMD SEV on C3D machine type don't
    support Hyperdisk Balanced and Hyperdisk Throughput.

### AMD SEV-SNP

- The AMD Secure Processor is a shared resource available to all hosts on
  the physical server it's present in. A host can throttle guest access to
  the Secure Processor, and firmware updates to the Secure Processor (which
  are considered host events) can happen at any time.

- Debian 12 doesn't have attestation support for AMD SEV-SNP due to the
  missing `/dev/sev-guest` package.

- AMD SEV-SNP on N2D machine types has a maximum
  [vNIC queue count](https://docs.cloud.google.com/compute/docs/network-bandwidth#rx-tx) of `8`.

- AMD SEV-SNP VM instances don't support `kdump`. Instead, use the guest
  console logs.

- Confidential VM instances with AMD SEV-SNP don't support
  [reservations](https://docs.cloud.google.com/compute/docs/instances/reservations-overview).

- Due to a guest kernel migration and security updates, Confidential VM
  instances with AMD SEV-SNP might have longer boot times and performance
  changes between August 2026 and November 2026.

### Intel TDX

- Intel TDX supports up to 192 vCPUs.

- [Local SSD](https://docs.cloud.google.com/compute/docs/machine-resource#lssd-types) is only supported
  on `c3-standard-*-lssd` machine types.

- Intel TDX VM instances take longer to shut down compared to standard VM
  instances. This delay increases with VM memory size.

- Only Balanced Persistent Disk volumes that use the NVMe interface are
  supported.

- Intel TDX VM instances can't be provisioned on sole-tenant node groups.

- Due to additional security constraints, the CPUID instruction might return
  limited or no CPU architecture details. This might impact the performance
  of workloads that depend on those CPUID values.

- Intel TDX VM instances don't support `kdump`. Instead, use the guest
  console logs.

- Guest images without the
  [TDX halt fixes](https://lore.kernel.org/lkml/20250228014416.3925664-1-vannapurve@google.com/)
  might experience extended halt durations, resulting in performance
  degradation. To avoid the performance degradation, verify that these
  patches are installed in your guest kernel build.

- Intel TDX VM instances don't support
  [reservations](https://docs.cloud.google.com/compute/docs/instances/reservations-overview).

### NVIDIA Confidential Computing

- GPU attestation might fail, reporting a mismatched measurement record at
  index 9. Google and NVIDIA are working on a fix.

  To work around the issue, perform a full
  [stop and restart of the Confidential VM](https://docs.cloud.google.com/compute/docs/instances/stop-start-instance).
  Rebooting the VM from the VM's guest OS does not fix the issue.
- GPU-accelerated VM instances don't support the creation of clusters for
  multi-node workloads.

- GPU-accelerated VM instances don't support
  [Hyperdisk Extreme](https://docs.cloud.google.com/compute/docs/disks/hd-types/hyperdisk-extreme).

- GPU-accelerated VM instances running on the
  [A3 High machine type](https://docs.cloud.google.com/compute/docs/accelerator-optimized-machines#a3-vms)
  with NVIDIA H100 GPUs don't support
  [reservations](https://docs.cloud.google.com/compute/docs/instances/reservations-overview).

- Machine type limitations apply. See the limitations for each
  machine type:

  - [A3 High](https://docs.cloud.google.com/compute/docs/accelerator-optimized-machines#a3-high-limitations)

  - [G4](https://docs.cloud.google.com/compute/docs/accelerator-optimized-machines#g4-limitations)

## Machine types

Confidential VM is supported on the following machine types.

| Machine type | CPU platform | GPU | Confidential Computing technology | Live migration support |
|---|---|---|---|---|
| `a3-highgpu-1g` [View machine type names](https://docs.cloud.google.com/compute/docs/accelerator-optimized-machines#a3-standard-vms) | - Intel Sapphire Rapids | NVIDIA H100 | - Intel TDX [See limitations](https://docs.cloud.google.com/confidential-computing/confidential-vm/docs/supported-configurations?tab=intel-tdx#limitations) - NVIDIA [See limitations](https://docs.cloud.google.com/confidential-computing/confidential-vm/docs/supported-configurations?tab=nvidia-confidential-computing#limitations) | Not supported |
| C2D [View machine type names](https://docs.cloud.google.com/compute/docs/compute-optimized-machines#c2d_machine_types) | - AMD EPYC Milan | None | - AMD SEV [See limitations](https://docs.cloud.google.com/confidential-computing/confidential-vm/docs/supported-configurations?tab=amd-sev#limitations) | Not supported |
| `c3-standard-*` [View machine type names](https://docs.cloud.google.com/compute/docs/general-purpose-machines#c3_machine_types) | - Intel Sapphire Rapids | None | - Intel TDX [See limitations](https://docs.cloud.google.com/confidential-computing/confidential-vm/docs/supported-configurations?tab=intel-tdx#limitations) | Not supported |
| `c3-standard-*-lssd` [View machine type names](https://docs.cloud.google.com/compute/docs/general-purpose-machines#c3-standard_1) | - Intel Sapphire Rapids | None | - Intel TDX [See limitations](https://docs.cloud.google.com/confidential-computing/confidential-vm/docs/supported-configurations?tab=intel-tdx#limitations) | Not supported |
| C3D [View machine type names](https://docs.cloud.google.com/compute/docs/general-purpose-machines#c3d_machine_types) | - AMD EPYC Genoa | None | - AMD SEV [See limitations](https://docs.cloud.google.com/confidential-computing/confidential-vm/docs/supported-configurations?tab=amd-sev#limitations) | Supported |
| `c4-standard-*` [View machine type names](https://docs.cloud.google.com/compute/docs/general-purpose-machines#c4_machine_types) | - Intel Granite Rapids | None | - Intel TDX [See limitations](https://docs.cloud.google.com/confidential-computing/confidential-vm/docs/supported-configurations?tab=intel-tdx#limitations) | Not supported |
| C4D [View machine type names](https://docs.cloud.google.com/compute/docs/general-purpose-machines#c4d_machine_types) | - AMD EPYC Turin | None | - AMD SEV [See limitations](https://docs.cloud.google.com/confidential-computing/confidential-vm/docs/supported-configurations?tab=amd-sev#limitations) | Not supported |
| `g4-standard-48` [View machine type names](https://docs.cloud.google.com/compute/docs/accelerator-optimized-machines#g4-series) | - AMD EPYC Turin | NVIDIA RTX Pro 6000 | - AMD SEV [See limitations](https://docs.cloud.google.com/confidential-computing/confidential-vm/docs/supported-configurations?tab=amd-sev#limitations) - NVIDIA [See limitations](https://docs.cloud.google.com/confidential-computing/confidential-vm/docs/supported-configurations?tab=nvidia-confidential-computing#limitations) | Not supported |
| N2D [View machine type names](https://docs.cloud.google.com/compute/docs/general-purpose-machines#n2d_machine_types) | - AMD EPYC Milan | None | - AMD SEV on Milan [See limitations](https://docs.cloud.google.com/confidential-computing/confidential-vm/docs/supported-configurations?tab=amd-sev#limitations) - AMD SEV-SNP on Milan [See limitations](https://docs.cloud.google.com/confidential-computing/confidential-vm/docs/supported-configurations?tab=amd-sev-snp#limitations) | Supported on AMD SEV VM instances only |

## Zones

You can view which zones support these machine types and Confidential Computing
technology with one of the following methods.

### AMD SEV

To list the zones that AMD SEV is supported in, run the following command:

    gcloud compute zones list \
        --filter="availableCpuPlatforms.list():'CPU_PLATFORM'" \
        --format="value(name)" \
        --sort-by=name

Replace `CPU_PLATFORM` with the name of the CPU platform
that you want to search for:

- `AMD Milan`

- `AMD Genoa`

- `AMD Turin`

### AMD SEV-SNP

AMD SEV-SNP is supported in the following zones, on N2D machine type with
AMD Milan CPU platforms:

- asia-southeast1-a
- asia-southeast1-b
- asia-southeast1-c
- europe-west3-a
- europe-west3-b
- europe-west3-c
- europe-west4-a
- europe-west4-b
- europe-west4-c
- us-central1-a
- us-central1-b
- us-central1-c

### Intel TDX

### c3-standard-\*

Intel TDX is supported in the following zones, on `c3-standard-*`
machine types.

- asia-northeast1-b
- asia-south1-b
- asia-southeast1-a
- asia-southeast1-b
- asia-southeast1-c
- europe-west3-a
- europe-west3-b
- europe-west4-a
- europe-west4-b
- europe-west4-c
- europe-west9-a
- europe-west9-b
- us-central1-a
- us-central1-b
- us-central1-c
- us-east1-c
- us-east1-d
- us-east4-a
- us-east4-b
- us-east4-c
- us-east5-b
- us-east5-c
- us-west1-a
- us-west1-b

### c3-standard-\*-lssd

Intel TDX is supported on `c3-standard-*-lssd` machine
types in the following zones:

- asia-southeast1-a
- asia-southeast1-b
- europe-west4-a
- europe-west4-b
- southamerica-east1-a
- us-central1-a
- us-central1-b
- us-east4-b
- us-east4-c
- us-east5-b
- us-east5-c
- us-west1-b

### c4-standard-\*

Intel TDX is supported in the following zones, on `c4-standard-*`
machine types:

- asia-southeast1-a
- asia-southeast1-b
- asia-southeast1-c
- europe-west1-b
- europe-west3-a
- europe-west3-b
- europe-west4-a
- europe-west4-b
- us-central1-a
- us-central1-b
- us-central1-f
- us-east4-a
- us-east4-b
- us-east4-c
- us-west1-a

### NVIDIA Confidential Computing

### A3 High

NVIDIA Confidential Computing is supported in the following zones, on Confidential VM
instances with an attached GPU on the A3 High machine type:

- europe-west4-c
- us-central1-a
- us-east5-a

### G4

To list the zones the `g4-standard-48` machine type is available in, run
the following command:

    gcloud compute accelerator-types list \
        --filter="name ~ ^nvidia-rtx-pro-6000$" \
        --format="value(zone)" \
        --sort-by=zone

## Operating systems

For the available Confidential VM operating system images, see
[Operating system details](https://docs.cloud.google.com/compute/docs/images/os-details). Find your
distribution of choice, and then click the **Security features** tab to check if
Confidential VM is supported.

Alternatively, you can
[view supported operating system images with a `gcloud` command](https://docs.cloud.google.com/confidential-computing/confidential-vm/docs/supported-configurations#supported-os),
or
[create your own Linux image](https://docs.cloud.google.com/confidential-computing/confidential-vm/docs/create-custom-confidential-vm-images).

### View supported operating system images with gcloud

The operating system images you can use are determined by your
Confidential Computing technology choice.

You can list the operating system images, their image families, and their
versions that support AMD and Intel Confidential Computing technologies by
running the following command:

    gcloud compute images list \
        --filter="guestOsFeatures[].type:(OS_FEATURE)"

Provide the following value:

`OS_FEATURE`: The type of Confidential Computing support you
want. Accepted values are:

- `SEV_CAPABLE`: Operating systems that support AMD SEV.

- `SEV_LIVE_MIGRATABLE_V2`: Operating systems that support AMD SEV and live
  migration.

- `SEV_SNP_CAPABLE`: Operating systems that support AMD SEV-SNP isolation and
  attestation.

- `TDX_CAPABLE`: Operating systems that support Intel TDX isolation and
  attestation.

To limit the results to a specific image family, project, or other text provided
in the previous command's response, use an `AND` operator, and replace
<var translate="no">STRING</var> with a partial text match, similar to the following example:

    gcloud compute images list \
        --filter="guestOsFeatures[].type:(OS_FEATURE) AND STRING"

To view details about a specific image, run the following command using details
from the responses of the previous commands:

    gcloud compute images describe IMAGE_NAME \
        --project=IMAGE_PROJECT

### Supported images for a Confidential VM instance with GPU

For Confidential VM instances with attached NVIDIA GPUs, the requirements for
operating system image projects and families are as follows:

- **A3 High machine type**:

  - **Container-Optimized OS** (`confidential-vm-images` project):
    `cos-121-lts` and newer

  - **Ubuntu** (`ubuntu-os-cloud` project): `ubuntu-2204-lts` and newer

  While other images might be tagged as `TDX_CAPABLE`, we don't provide official
  support for them when used with GPUs.
- **G4 machine type**:

  - **Container-Optimized OS** (`cos-cloud` project): `cos-125-lts`
    and newer

  - **Ubuntu** (`ubuntu-os-cloud` project):

    - `ubuntu-2204-lts` and newer

    - `ubuntu-2404-lts-amd64` and newer

    - `ubuntu-minimal-2204-lts` and newer

    - `ubuntu-minimal-2404-lts-amd64` and newer

  - **Ubuntu Pro** (`ubuntu-os-pro-cloud` project):

    - `ubuntu-pro-2204-lts` and newer

    - `ubuntu-pro-2404-lts-amd64` and newer

### Supported images for C3 with Local SSD

For Confidential VM instances using Intel TDX on `c3-standard-*-lssd` machine
types, the following operating system image families are supported:

- `cos-117-lts`

- `cos-121-lts`

## What's next

- Learn how to
  [create a Confidential VM instance](https://docs.cloud.google.com/confidential-computing/confidential-vm/docs/create-a-confidential-vm-instance).

- Learn how to
  [create a Confidential VM instance with GPU](https://docs.cloud.google.com/confidential-computing/confidential-vm/docs/create-a-confidential-vm-instance-with-gpu).