Every Linux VM on Proxmox mostly just works. Pick a template, boot it, maybe adjust the network config, done. Windows is a different animal, not because Proxmox handles it worse, but because Windows makes assumptions about the hardware underneath it that KVM’s default virtual machine doesn’t match out of the box. Skip the setup steps below and you’ll either hit a wall during install (no TPM, Windows 11 refuses to continue) or end up with a VM that technically boots but crawls on disk and network I/O because it’s using emulated hardware instead of VirtIO. Here’s the order that actually works, and why each step exists.

Machine type and BIOS: q35 and OVMF, not the defaults

Proxmox’s default new-VM wizard still offers the older i440fx machine type and SeaBIOS, which is fine for a lot of Linux guests but wrong for Windows if you want UEFI, Secure Boot, or TPM 2.0 support (Windows 11 requires all three to install cleanly, and Windows 10 benefits from them too).

  • Machine type: q35. This emulates a more modern PCIe-based chipset instead of the older PCI-based i440fx. Set it when creating the VM, under the System tab, or in the Hardware tab’s Machine field for an existing one. Changing an already-installed VM’s machine type later can shuffle its PCI device addressing enough to upset Windows, so decide this before you install, not after.
  • BIOS: OVMF (UEFI), not SeaBIOS. SeaBIOS is legacy BIOS emulation and simply can’t present a TPM or do Secure Boot, full stop. Switch this on the System tab when creating the VM.
  • Add an EFI disk. Selecting OVMF adds a small EFI disk automatically in the wizard, make sure it’s actually there if you’re editing an existing VM’s config by hand, since UEFI firmware needs somewhere to persist its boot variables, and losing that disk means losing the boot entry.

The TPM state disk: the step everyone forgets

Windows 11’s install check (and some Windows 10 security features) looks for a TPM 2.0 chip. Your Proxmox host almost certainly doesn’t have a physical TPM wired up to guests, so Proxmox emulates one through swtpm, but it needs its own small virtual disk to store the TPM’s state, separate from the EFI disk above.

In the VM’s Hardware tab, Add > TPM State, pick version v2.0 (not v1.2, Windows 11 specifically wants 2.0), and pick any storage that supports it. This single checkbox is the difference between a Windows 11 ISO booting into setup normally and bailing out at “This PC can’t run Windows 11” before you’ve typed anything. If you’re migrating an existing Windows VM that was built without one, you can add the TPM state disk after the fact, but you still need to have Secure Boot and UEFI already in place for it to actually satisfy the check, a bare TPM device with SeaBIOS still won’t pass.

VirtIO drivers: the actual performance unlock

By default, a freshly created VM in Proxmox uses the SCSI controller type and network card model that are safest for broad compatibility, not the fastest. For Windows specifically, that means your install disk shows up as a generic SCSI or SATA device and your network adapter shows up as an emulated Intel or Realtek NIC, both of which work, but both of which leave a meaningful amount of performance on the table compared to VirtIO’s paravirtualized drivers.

The catch: Windows doesn’t ship with VirtIO drivers built in, so if you set the disk controller to VirtIO SCSI before installing, the Windows installer won’t see any disk to install onto at all, because it has no driver for it yet. The fix is a second virtual CD/DVD drive:

  1. Attach the Windows installer ISO as the primary CD/DVD drive.
  2. Attach the VirtIO driver ISO (the official virtio-win package, downloadable from the Fedora-hosted VirtIO driver project) as a second CD/DVD drive.
  3. Set the VM’s hard disk controller to VirtIO SCSI single and the network card to VirtIO (paravirtualized) before you ever boot the installer.
  4. During the Windows install, when it asks where to install Windows and shows no drives, click Load driver, point it at the VirtIO ISO’s viostor (or vioscsi for the “single” controller variant) folder matching your Windows version, and the disk will appear.
  5. After Windows is installed and booted, run the VirtIO driver ISO’s installer from inside the guest to pick up the rest of the driver set: network, balloon memory driver, serial port, and the guest tools. Doing this from inside Windows after first boot is simpler than hunting down every individual .inf during setup, and it’s the step people most often skip entirely, leaving the NIC on the slower emulated driver indefinitely.

The balloon driver and memory, briefly

The VirtIO balloon driver is what lets Proxmox reclaim idle RAM from a guest and hand it to other VMs under memory pressure, the same mechanism covered in more depth elsewhere on this site for Linux guests. It behaves the same way on Windows once installed from the VirtIO ISO, but Windows is noticeably worse than Linux about actually releasing memory back under pressure, its own memory manager tends to hold onto what it’s been given. If you’re running ballooning on a Windows VM, don’t be surprised if the guest sits closer to its configured maximum than a Linux VM would under the same settings, that’s normal Windows behavior, not a broken balloon driver.

qemu-guest-agent: don’t skip this either

Install the QEMU guest agent (included in the same virtio-win ISO, or installable separately) and enable it in the VM’s Options tab (QEMU Guest Agent: Enabled). Without it, Proxmox has no clean way to tell the guest to shut down (a “shutdown” from the Proxmox UI falls back to an ACPI power event the OS may or may not honor cleanly), can’t report the guest’s actual IP address in the summary pane, and backup jobs can’t ask the guest to quiesce its filesystem before the snapshot is taken, which matters more than it sounds for a VM running a database or anything else sensitive to being backed up mid-write.

Secure Boot, if you’re turning it on

If you enable Secure Boot (OVMF supports it, there’s a toggle on the Hardware tab’s BIOS/EFI settings), do it before installing Windows, not after, since Windows’s boot loader gets signed and registered against Secure Boot’s key database at install time. Turning Secure Boot on for an existing, already-installed VM will usually just fail to boot until you either turn it back off or go through Windows’s own Secure Boot key management, neither of which is worth the hassle for a homelab VM that wasn’t installed with it from the start.

CPU type: don’t leave it on the default either

The default CPU type (kvm64 or similar) emulates a lowest-common-denominator CPU for maximum migration compatibility across different host hardware, at the cost of hiding newer instruction sets from the guest. For a single-host homelab where you’re not live-migrating this VM to different physical hardware, set the CPU type to host instead, it passes the actual host CPU’s full feature set through to the guest, which matters for anything CPU-intensive inside the VM (games, video encoding, anything using AVX or similar). The tradeoff, covered in more depth in the live migration article elsewhere on this site, is that host CPU type VMs can’t live-migrate to a host with a different CPU, which is irrelevant on a single Proxmox box anyway.

Bottom line

The order matters: pick q35 and OVMF before you create the disk, add the TPM state disk before you boot the installer, attach the VirtIO driver ISO alongside the Windows ISO so the installer can actually see your disk, and install the guest agent and full VirtIO driver set from inside Windows right after first boot, not months later when you notice the NIC feels slow. Get those five things right in that order and a Windows VM on Proxmox performs and behaves close to what you’d get on bare metal. Skip the TPM disk or the VirtIO ISO and you’ll either not get through setup at all, or end up running Windows on emulated hardware indefinitely without realizing there was a faster path available the whole time.