OPNsense

Home›Blog›Virtual Machines›OPNsense on Proxmox: Bridges, VirtIO and PCI Passthrough Compared

OPNsense on Proxmox: Bridges, VirtIO and PCI Passthrough Compared

Category: Virtual Machines · Updated 2026-11-19 · Applies to OPNsense 26.7

Illustration of virtual network bridges connecting a firewall VM to physical ports

When you run OPNsense on Proxmox VE, you have three real choices for connecting the firewall to your networks: virtual network cards on Linux bridges using the paravirtualised VirtIO driver, virtual cards on the same bridges using an emulated Intel adapter, or PCI passthrough, which hands a physical network card to the VM directly. Each option trades performance, compatibility and flexibility differently, and the right answer depends on your hardware and how fast your connection is. This guide compares all three honestly, then covers the two things that matter regardless of your choice: disabling hardware offloading in OPNsense, and keeping Proxmox management reachable even when the firewall VM is down.

The building block: Linux bridges

A Proxmox Linux bridge (named vmbr0, vmbr1 and so on) is a software switch. You bind a physical network port to the bridge, and every virtual network card attached to that bridge behaves as if it were plugged into the same switch as that port. For a firewall VM you normally use two bridges: one whose physical port faces your modem or upstream router (the WAN side), and one whose port faces your local switch (the LAN side). Bridges can also be made VLAN-aware, so a single physical port can carry several tagged networks at once, which is how single-port hosts separate WAN and LAN through a managed switch.

With bridged networking, Proxmox itself moves packets between the VM and the physical port. That indirection is what makes bridges flexible: you can attach several VMs to the same bridge, migrate configurations easily, and change cabling without touching the VM. The cost is some CPU overhead on the host and one extra layer to troubleshoot. The two bridged options below differ only in the kind of virtual network card the VM sees.

Option 1: Bridged VirtIO network cards

VirtIO is the paravirtualised device family designed for KVM, the hypervisor Proxmox is built on. Instead of pretending to be a piece of physical hardware, a VirtIO network card knows it is virtual and talks to the hypervisor through an efficient shared-memory interface. FreeBSD, and therefore OPNsense, ships with VirtIO drivers, so the cards work out of the box and appear as vtnet interfaces.

In practice, VirtIO on bridges is the default choice for most Proxmox firewall builds, and it is what the VM created in How to Install OPNsense on Proxmox VE uses. It offers the best throughput of the two bridged options with the lowest host CPU cost, because packets take a shorter path than they do through a hardware emulation layer. One honest caveat: the official OPNsense VM documentation covers KVM chipsets, the guest agent and offloading settings, but it does not publish a blanket "always use VirtIO" instruction, so treat VirtIO as well-established community practice rather than an official mandate. The one related setting the official guide does insist on, disabling hardware offloading, applies here and is covered below.

Option 2: Bridged emulated Intel (e1000) cards

Proxmox can also present the VM with an emulated Intel network adapter, commonly the e1000. The hypervisor simulates the hardware register by register, so almost any guest operating system recognises it with built-in drivers. OPNsense sees it as an em interface, just like real Intel hardware.

The trade-off is performance. Emulation is expensive: every packet crosses extra layers, host CPU usage rises, and throughput tops out well below what VirtIO achieves on the same hardware. For a slow internet connection and a light workload, you may never notice. For gigabit routing, VPN traffic or inter-VLAN routing at speed, the emulated adapter is usually the bottleneck. Keep this option in your pocket as a compatibility fallback: if a VirtIO interface ever misbehaves after an update, switching the card model to e1000 is a quick way to separate a driver problem from a configuration problem. As a permanent design, it rarely makes sense on KVM.

Option 3: PCI passthrough of a physical NIC

PCI passthrough removes the virtual switch entirely. Using the host's IOMMU, you assign a whole physical network card (or one port of a multi-port card, if the hardware separates ports into their own IOMMU groups) directly to the VM. OPNsense then drives the real hardware with its native FreeBSD driver, exactly as it would on a physical install.

The advantages are real: the lowest latency, no bridge overhead, and behaviour identical to bare metal, including full use of the card's own features. The disadvantages are equally real:

  • The card disappears from the host. Proxmox cannot use that port for itself or for any other VM.
  • Your motherboard and CPU must support IOMMU (Intel VT-d or AMD-Vi), and it must be enabled in firmware and in Proxmox.
  • The VM becomes tied to this host and this hardware, so live migration and easy hardware swaps are off the table.
  • Troubleshooting changes character: a driver or hardware quirk is now a firewall problem, not a hypervisor problem.

Passthrough is a good fit when you have a spare multi-port card, a fast connection, and a host that does little else. It is a poor fit on a single-port host, where passing through the only NIC would cut the host off completely.

Side-by-side comparison

CriterionBridge + VirtIOBridge + e1000PCI passthrough
Throughput potentialHighModestNative, highest
Host CPU costLowHighLowest (no switching)
Hardware requirementsNone specialNone specialIOMMU support, spare NIC
FlexibilityExcellent: share bridges, rewire in softwareExcellentPoor: card is dedicated to the VM
Driver in OPNsenseVirtIO (vtnet)Emulated Intel (em)Native driver for the real card
Typical useDefault for most buildsCompatibility fallbackDedicated firewall hosts, fast links

The setting that applies to every option: disable offloading

Whichever model you choose, open the OPNsense web GUI, go to Interfaces → Settings, and disable all hardware offloading (checksum offloading, TCP segmentation offloading and large receive offloading), then reboot. The official OPNsense VM guide recommends exactly this for virtual machines, because offload features assume a direct relationship between the driver and physical hardware that virtual and emulated cards cannot honour. Symptoms of leaving it on range from disappointing throughput to connections that stall in ways that look like firewall rule problems. It is also a hard prerequisite if you later enable IPS mode, which will not function with offloading enabled.

VLANs on bridges: one port, several networks

Bridges have one more trick that matters for firewall design. A Proxmox bridge can be marked VLAN-aware, after which a VM network card can be given a VLAN tag, and the bridge will deliver only that VLAN's traffic to the card while the physical port carries all the tagged networks at once. This is how a host with a single physical port can still separate WAN and LAN, or offer OPNsense several internal segments (main LAN, guests, IoT) over one cable to a managed switch.

It works well, but it moves complexity into configuration. The tag on the VM card, the trunk settings on the switch port, and the VLAN interfaces inside OPNsense must all agree, and a mismatch fails silently into "no traffic" rather than an error message. Two pieces of official guidance are worth following here. First, do not mix tagged and untagged traffic on the trunk between OPNsense and the switch; keep the trunk purely tagged. Second, if you are connecting two ports to the same switch for redundancy or bandwidth, use a LAGG (link aggregation) rather than a bridge, because bridging two paths to the same switch invites loops. For most homes, one untagged port per network stays the design to beat: it is the easiest to reason about at 2 a.m. when something breaks.

Changing your mind later

One comfort of bridged networking is that the choice is not permanent. You can shut the VM down, change a network card's model from VirtIO to e1000 (or back), or move it to a different bridge, all from the Proxmox hardware tab. Be aware of the side effect inside OPNsense: a different card model means a different driver and therefore a different interface name (vtnet0 becomes em0, for example), so OPNsense will ask you to reassign interfaces on the next boot, and any configuration tied to the old names (VLANs, additional interfaces) needs to be re-pointed. Passthrough is the exception to this flexibility: adding or removing a passed-through card is a bigger operation, and the interface appears with the native driver's name. In all cases, take a configuration backup (System → Configuration → Backups) before you start swapping virtual hardware.

The host-management trap

The most dangerous Proxmox firewall mistake has nothing to do with drivers. It happens when the Proxmox host's own management address sits on a network that only works while the firewall VM is running. Picture it: the host's management interface is on the LAN bridge, the LAN only gets DHCP, DNS and its gateway from OPNsense, and the VM fails to boot after an update. Now you cannot reach the web interface of the firewall (it is down) or of Proxmox (its network path ran through the firewall's LAN). You are reduced to a physical keyboard on the host.

Avoid it with one of these designs:

  • Give Proxmox a static management address on the LAN bridge that does not depend on the VM for its own connectivity, and make sure your workstation can reach that address directly on the local switch, firewall or no firewall.
  • Better still, put host management on its own port or bridge that never passes through the VM.
  • Keep a fallback: a monitor and keyboard for the host, or a remote management interface (IPMI, iDRAC, Intel AMT or similar) if your hardware has one.
  • Set the firewall VM to start at boot, and delay dependent VMs until it is up.

The rule of thumb: you should always be able to reach the Proxmox console with the firewall VM powered off. Test this deliberately once, by shutting the VM down and opening the Proxmox interface, before you rely on the setup.

How to confirm your choice is working

  • In OPNsense under Interfaces → Assignments, each interface's MAC address matches the card and bridge you intended in Proxmox (or, with passthrough, the physical port you assigned).
  • The WAN interface holds the expected address (DHCP from your modem or router, or your static assignment), and the LAN interface shows your LAN address, 192.168.1.1/24 by default.
  • A LAN client can reach the internet, and a simple speed test lands in the range your internet plan and your sizing suggest; wildly low results with high host CPU point at emulation overhead or offloading.
  • Interfaces → Settings shows every hardware offloading option disabled.
  • With the firewall VM shut down, you can still open the Proxmox web interface from your workstation.

A brief note on machine type

While you are choosing network cards, Proxmox will also ask (or will have chosen for you) a machine type for the VM: the legacy i440FX or the newer Q35 chipset emulation. For OPNsense this is not a performance decision. The official VM documentation notes that OPNsense, through FreeBSD, supports both machine types on KVM, and your network cards work the same on either. Q35 is the more modern platform and the natural choice for a new VM; i440FX remains a perfectly good fallback if a particular host or an older configuration prefers it. Pick one, note that interface names and assignments are unaffected by the choice, and spend your attention on the bridges and card models instead, because those are what actually shape how the firewall behaves.

Which should you pick?

Start with bridges and VirtIO. It is flexible, fast enough for most home and small-office connections, and easy to change later. Choose PCI passthrough when you have spare ports, IOMMU support, and a reason to squeeze out native performance or bare-metal behaviour. Use the emulated Intel adapter only as a diagnostic step or a short-term fallback. Whichever you choose, size the VM properly using OPNsense Virtual Machine Requirements and Sizing, and finish the job with the QEMU Guest Agent so shutdowns and snapshots behave properly. The official background for this topic lives in the OPNsense virtual machines documentation.

Keep going in Virtual Machines