Advanced Guide: QEMU CPU Models for KVM Guests - Debunking the "host" CPU Model Performance Myth

Note

Target single-node homelab setups using AVX2-capable processors.

Tip

Always install the latest CPU microcode for your CPU.
Details: https://pve.proxmox.com/pve-docs/pve-admin-guide.html#sysadmin_firmware_cpu

References

• QEMU x86 CPU Models Documentation
• QEMU Hyper-V Enlightenments Documentation

---

Simple Windows Guests

For basic Windows guests without WSL2, Hyper-V, or VBS enabled, use named QEMU CPU models corresponding to your hardware. For example, if you have Skylake CPUs, use Skylake-Client-v4.

Note

Using the default host model in Proxmox VE will result in the following QEMU -cpu argument when starting the Windows VM:

qm show VMID --pretty

-cpu 'host,hv_ipi,hv_relaxed,hv_reset,hv_runtime,hv_spinlocks=0x1fff,hv_stimer,hv_synic,hv_time,hv_vapic,hv_vpindex,+kvm_pv_eoi,+kvm_pv_unhalt'

This is fine as long as you don't have any of these enabled in your Windows VM: WSL2, Hyper-V, or VBS.

---

Hyper-V, WSL2, and VBS Enabled Guests

To achieve the best performance with Hyper-V, WSL2, and VBS enabled, we need to create a custom CPU model:

1. Create the file /etc/pve/virtual-guest/cpu-models.conf with the following content:

# Proxmox VE Custom CPU Models
cpu-model: amd-hide-vm-for-windows
    flags -hypervisor;+invtsc;+hv-frequencies;+hv-reenlightenment;+hv-emsr-bitmap;+hv-tlbflush-direct
    phys-bits host
    hidden 1
    hv-vendor-id amd
    reported-model host

cpu-model: intel-hide-vm-for-windows
    flags -hypervisor;+invtsc;+hv-frequencies;+hv-evmcs;+hv-reenlightenment;+hv-emsr-bitmap;+hv-tlbflush-direct
    phys-bits host
    hidden 1
    hv-vendor-id intel
    reported-model host

2. Select either amd-hide-vm-for-windows or intel-hide-vm-for-windows CPU model in the Proxmox web GUI, depending on your processor.

Note

Using these custom CPU models will result in the following QEMU -cpu argument when starting the Windows VM:

qm show VMID --pretty

-cpu 'host,+hv-emsr-bitmap,+hv-evmcs,+hv-frequencies,+hv-reenlightenment,+hv-tlbflush-direct,hv_ipi,hv_relaxed,hv_reset,hv_runtime,hv_spinlocks=0x1fff,hv_stimer,hv_synic,hv_time,hv_vapic,hv_vendor_id=intel,hv_vpindex,-hypervisor,+invtsc,kvm=off,+kvm_pv_eoi,+kvm_pv_unhalt,host-phys-bits=true'

Tip

With these custom CPU models configured, you can now run Android emulators very smoothly inside your Windows VM. You need to have a GPU passthrough to the VM and enable Hyper-V:

Enable-WindowsOptionalFeature -Online -FeatureName Microsoft-Hyper-V -All

Tested with MEmu Play.

---

macOS Guests

Performance Considerations

• Always use named QEMU CPU models instead of host passthrough
• The host passthrough can result in ~30% slower single-core and ~44% slower multi-core performance compared to named QEMU CPU models

CPUID Model Compatibility

The macOS kernel checks CPUID model and vendor during boot. When using CPU models like SapphireRapids or GraniteRapids (which were never shipped in real Macs), you must override the CPUID model to a known Mac-compatible model. While this can be configured in OpenCore's config.plist (Kernel > Emulate Properties), QEMU-level configuration is recommended.

Tip

CPUID Models Used in Recent Intel Macs:

Generation	CPUID Model
Kaby Lake / Coffee Lake	158
Comet Lake	165

Example Configuration:

qm set <VMID> --args "-cpu SapphireRapids,vendor=GenuineIntel,model=165"

---

Bonus: Advanced CPU Spoofing (cosmetic purpose)

Customizing CPU Brand Name

When using named QEMU CPU models like Skylake-Client-v4, the guest OS will report the CPU as Intel Core Processor (Skylake, IBRS, no TSX). You can customize this using the model-id parameter.

Spoofing CPU Frequency

For modern CPUs, you can spoof the reported base speed using the tsc-frequency parameter.

Example: Complete CPU Spoofing

To spoof a 10GHz CPU with the brand name "Intel Core i8-8800KS":

# Using global parameters
qm set <VMID> --args "-global cpu.model-id='Intel Core i8-8800KS' -global cpu.tsc-frequency=10000000000"

Additional Parameters

Many more CPU parameters are available, including:

• family= and model= - CPU family and model numbers
• stepping= and level= - CPU stepping and CPUID level
• host-cache-info=on - Pass through host cache information

For complete details, see: https://github.com/qemu/qemu/blob/master/target/i386/cpu.c