Instance Types & Preferences

Instance types and preferences provide predefined compute and hardware profiles for VMs. Use them to standardize sizing and reduce configuration errors.

VM Instance Types

VirtualMachineClusterInstancetype resources define CPU and memory presets. They are cluster-scoped and available in all namespaces.

What VM Instance Types Define

Field	Description
CPU	Guest-visible vCPU count.
Memory	Guest-visible memory, such as 2Gi or 4Gi.
CPU Model	Optional. The named x86 CPU model exposed to the guest, such as Skylake-Server or EPYC-Rome, or a passthrough mode such as host-model or host-passthrough.

VM instance types do not define storage, network, or other hardware. Those come from the VM spec, template, or preference.

Built-in and Custom Instance Types

PaletteAI VM Launchpad and KubeVirt ship with example instance types, such as u1.small, u1.medium, and cx1.medium. Platform Admins can add custom instance types on the Workloads > Instance Types page. Cluster administrators can also create them through the Custom Resource Definition (CRD).

Prefix Conventions

Instance types are often grouped by a naming prefix.

Prefix	Typical Use
u1	General purpose (balanced CPU/memory)
cx1	Compute optimized (more CPU)
mx1	Memory optimized (more memory)

The VM creation wizard displays instance types in a FilterableCardSelector with collapsible groups. You can filter by name or spec. For example, search for 4 to find 4 vCPU types.

CPU Model Field

The CPU Model dropdown in the Create/Edit Instance Type modal is populated dynamically from the cluster. It lists only the models that at least one schedulable worker node reports as supported, as discovered from KubeVirt's cpu-model-migration.node.kubevirt.io/<Model> node labels (written by KubeVirt's node-labeller based on each node's physical CPU features).

Two warnings can appear inline under the select:

• This CPU model is not available on any cluster node. VMs may fail to schedule. The selected model is not supported by any node in the cluster. This warning commonly appears after importing an instance type or template from another cluster with different CPU hardware. Pick a supported model from the dropdown, or add a node that exposes the required CPU features.
• This model is not supported by all nodes. Live migration may be restricted. The model is supported by some nodes but not all. VMs using this model can still schedule, but live migration between nodes is constrained to the subset that supports the model.

When an instance type or template references a model that the current cluster does not expose, the dropdown preserves the original value as <ModelName> (unsupported in current cluster) so importing does not silently drop the setting. You can keep the preserved value or choose a supported model before saving.

info

The list of discovered CPU models is cached for up to three minutes. The dropdown reflects an added, removed, or cordoned worker node within one cache window.

Instance Types in the Wizard

1. In the Compute step, select Instance Type (recommended).
2. Browse or filter the instance type cards. Each card shows CPU, memory, and other spec details.
3. Select an instance type. The SpecPreview shows the applied settings.
4. The generated VM spec uses spec.instancetype and omits explicit cpu and memory from the domain. The instance type provides those values.

Custom Mode vs Instance Type Mode

In the Compute step, you choose between two modes.

Mode	Behavior
Instance Type	Select a predefined profile. CPU and memory come from the instance type. Run strategy and eviction strategy are still configurable.
Custom	Set CPU cores and memory manually. Full control over resource requests, limits, and scheduling.

Instance type mode is recommended for consistency across teams. Custom mode is for workloads that need specific sizing not covered by instance types.

VM Preferences

VirtualMachineClusterPreference resources define OS-specific hardware and device defaults. They are cluster-scoped and filtered by guest OS.

What VM Preferences Define

Setting	Examples
Machine type	q35, pc-q35-rhel9.2.0
Firmware	EFI, Secure Boot, SMBIOS
Devices	RNG, tablet, video type, TPM
Features	ACPI, APIC, Hyper-V enlightenments (for Windows)
Clock	UTC versus timezone, timer configuration

VM preferences do not define CPU or memory. Those come from the instance type or custom compute settings.

Filter by OS

VM preferences are filtered by the Guest OS selected in the Source step.

Guest OS	Shows	Hides
Linux	Linux and "any" preferences	Windows-specific preferences
Windows	Windows and "any" preferences	Linux-specific preferences

This ensures you do not apply a Windows preference, such as Hyper-V, to a Linux VM or vice versa.

Preferences in the Wizard

1. In the Source step, set Guest OS (Linux or Windows).
2. Expand VM Preference (optional, collapsible section).
3. Browse or filter preference cards. Select a preference or leave it set to None.
4. The selected preference's settings are applied to the VM spec. The SpecPreview shows key settings, such as EFI, Hyper-V, and q35.

How VM Instance Types and VM Preferences Work Together

Component	Provides
VM Instance Type	CPU, memory
VM Preference	Machine type, firmware, devices, features
Template / Wizard	Storage, network, cloud-init, lifecycle

When both are selected:

• The VM spec includes spec.instancetype (for CPU/memory) and spec.preference (for hardware defaults).
• The domain is built by merging: instance type for resources, preference for devices/firmware, and explicit wizard choices for overrides.

Manage Instance Types and Preferences

Manage both resource types from the Workloads menu. They use the generic CRD resource UI, so you can list, create, edit, and delete them the same way. Changes apply to newly created VMs and templates. Existing VMs are not modified.

Only the Platform Admin VMO role can create, edit, or delete instance types and preferences. Other roles can view the lists but not change them, and Editors can still select instance types and preferences when they create VMs and templates.

Resource	Manage from	Example custom names
Instance types	Workloads > Instance Types	app-small, app-medium, db-large
Preferences	Workloads > Preferences	rhel9-secure, windows-2022-standard

Create a Custom Instance Type

Create a custom instance type when the built-in profiles do not match your sizing needs. This task requires the Platform Admin VMO role.

1. From the left main menu, select Workloads > Instance Types.

2. Select New Instance Type.

3. In the Name field, enter a name for the instance type, such as app-medium.

4. (Optional) To author the resource directly as YAML, select the </> toggle, edit the manifest, and select Apply YAML. Refer to YAML Drawer for details.

5. In the CPU section, configure the following settings. For field definitions, refer to What VM Instance Types Define and CPU Model Field.

   Setting	Description
   Guest vCPUs	The number of vCPUs exposed to the guest.
   CPU Model	Select Host Model (default) or host-passthrough.
   Max Sockets	Enables CPU hot plug when set to a value greater than Guest vCPUs. Defaults to auto. Refer to CPU Socket Changes for the runtime behavior when you change sockets.
   Dedicated CPU Placement	Pins the guest vCPUs to dedicated physical CPUs on the host.
   Isolate Emulator Thread	Isolates the QEMU emulator thread onto its own CPU. Helpful for jitter-sensitive applications.
   NUMA Guest Mapping Passthrough	Passes the host NUMA topology to the guest. Can improve performance for NUMA-aware applications.

6. In the Memory section, configure the following settings.

   Setting	Description
   Guest Memory	The memory exposed to the guest, in Mi, Gi, or Ti.
   Max Guest Memory	Enables memory hot plug when set to a value greater than Guest Memory.
   Hugepages Size	Select None (standard pages), 2Mi, or 1Gi. Larger pages can improve CPU performance for VMs with large Guest Memory.
   Overcommit %	The overcommit percentage applied on the hypervisor. A value of 0 applies no overcommit.

7. (Optional) In the Devices section, add devices.

   Device	Description
   GPUs	Map the VM to GPUs on the host.
   Host Devices	Add PCI passthrough or other host devices.

8. (Optional) In the Advanced section, configure the following settings.

   Setting	Description
   IO Threads Policy	Controls disk I/O isolation. Set to None (default), auto, shared, or supplementalPool. Refer to IO Thread Policy.
   AMD SEV (Confidential Computing)	Enables memory encryption.

9. (Optional) Add Labels as key-value pairs to organize the resource.

10. Select Create Instance Type to save the instance type.

The new instance type appears in the Instance Types list and becomes available in the Compute step of the VM creation wizard.

Create a Custom Preference

Create a custom preference to define OS-specific hardware and device defaults, such as rhel9-secure or windows-2022-standard. This task requires the Platform Admin VMO role.

1. From the left main menu, select Workloads > Preferences.

2. Select New Preference.

3. In the Name field, enter a name for the preference, such as rhel9-secure.

4. (Optional) To author the resource directly as YAML, select the </> toggle, edit the manifest, and select Apply YAML. Refer to YAML Drawer for details.

5. In the CPU section, configure the following settings.

   Setting	Description
   Topology	Sets how the guest vCPUs map to sockets, cores, and threads. Spread (default) distributes them evenly for the best NUMA performance. You can instead select Prefer Cores, Prefer Sockets, or Prefer Threads.
   Spread Across	When Topology is Spread, sets the units to spread vCPUs across: Sockets & Cores (recommended), Cores & Threads, or Sockets, Cores, & Threads.
   Spread Ratio	Sets the ratio of vCPUs per spread unit.

6. In the Devices section, configure the following settings.

   Setting	Description
   Disk Bus	virtio offers the best performance. Use sata or scsi for legacy disks.
   Disk Cache	writethrough is safest because it writes to the host and guest simultaneously. writeback improves performance but risks data loss if the host fails. none turns off the disk cache.
   Interface Model	Sets the network interface. virtio is the fastest. e1000 and e1000e provide broad support without virtio drivers.
   Hardware RNG	Enabled by default. Provides cryptographic entropy and is recommended for all VMs.

7. In the Machine & Firmware section, configure the following settings.

   Setting	Description
   Machine Type	q35 is a modern chipset with PCIe support. pc (i440fx) is a legacy chipset that supports PCI only.
   Termination Grace Period	The time, in seconds, to wait for a graceful shutdown before the VM is forced off.
   EFI Boot	Enabled by default. Uses UEFI firmware, which is required for Secure Boot and modern OS installers.
   Secure Boot	Enabled by default. Requires EFI Boot and SMM. Validates the boot chain to help prevent attacks at boot.

8. In the Features section, enable the options your guest OS requires.

   Feature	Description
   SMM	System Management Mode. Enabled by default with EFI Boot.
   ACPI	Advanced Configuration and Power Interface. Required for graceful shutdown on most guest operating systems.
   APIC	Advanced Programmable Interrupt Controller. Required for multi-core VMs.
   PV Spinlock	Reduces CPU overhead from lock contention on Linux guests through paravirtual spinlock optimization.
   Hyper-V Enlightenments	Windows-specific paravirtual optimizations that improve Windows guest performance.

9. In the Clock section, configure the following settings.

   Setting	Description
   UTC Clock Offset	Sets the guest hardware clock to UTC. Enabled by default and recommended for Linux. Turn it off for Windows guests, which expect local time.
   Windows Timers (HPET/PIT/RTC/Hyper-V)	Applies the standard Windows timer configuration.

10. In the Annotations section, the Descheduler eviction annotation is enabled by default. It marks the VM as eligible for descheduler-initiated live migration to improve cluster balance.

11. (Optional) Add Labels as key-value pairs to organize the resource.

12. Select Create Preference to save the preference.

The new preference appears in the Preferences list and becomes available in the Source step of the VM creation wizard.

YAML Drawer

The create and edit modals for both Instance Types and Preferences include a YAML drawer. Select the </> toggle to open it. The drawer shows the complete KubeVirt resource manifest, not just the spec, so you can author or inspect the full object.

apiVersion: instancetype.kubevirt.io/v1beta1
kind: VirtualMachineClusterInstancetype # or VirtualMachineClusterPreference
metadata:
  name: my-instance-type
  labels:
    tier: compute
spec:
  cpu:
    guest: 4
  memory:
    guest: 8Gi

Behavior

Field	Create mode	Edit mode
metadata.name	Syncs into the name field when you select Apply YAML.	Ignored because Kubernetes does not allow in-place renames.
metadata.labels	Applied to the form on Apply YAML.	Applied to the form on Apply YAML.
spec	Applied to the form on Apply YAML.	Applied to the form on Apply YAML.

If the YAML contains syntax errors, the editor catches them before running Apply. An error message appears in the drawer footer, and the form does not change.