Independent NICs or an Internal Switch? What ‘4× LAN’ Actually Means on an Industrial PC — and How to Verify It
Four RJ45 connectors on an industrial PC can be built two completely different ways, and the datasheet rarely says which:
- Four independent network controllers — four separate MACs, four PCIe lanes, four interfaces in the operating system. Each port can sit on its own subnet, carry its own protocol, and be assigned to its own trust zone.
- One controller behind an internal switch chip — four physical sockets, but one network interface in the operating system, one MAC address, one IP address, and shared bandwidth.
If your application needs separate subnets, OT/IT segmentation, an EtherCAT master, or deterministic fieldbus timing, only the first one works. The second is a built-in unmanaged switch, which is a useful thing — but it is not four network ports.
You can verify which one you have in about thirty seconds. The commands are in section 3.
1. Why the distinction decides your architecture
Separate subnets become impossible
This is the consequence most projects hit first. An internal switch presents one interface to the operating system. One interface takes one IP configuration. You cannot put LAN1 on 192.168.1.x for the device network and LAN2 on 10.0.0.x for the plant network, because the OS does not see LAN1 and LAN2 — it sees eth0.
Any architecture that depends on routing between two networks, or on keeping two networks apart, needs independent controllers.
OT/IT segmentation stops being physical
A switched block is a single broadcast domain unless the switch is managed and you can configure VLANs on it. Most embedded switch chips in industrial PCs are unmanaged — no VLAN tagging control, no port isolation, no mirroring.
The point of buying four ports for segmentation is that a configuration mistake cannot bridge two networks. On a switched block, they are already bridged in hardware.
EtherCAT and deterministic fieldbus break
An EtherCAT master takes exclusive control of a network interface, bypassing the operating system’s normal network stack to control transmission timing directly. That requires an interface it can own.
On a switched block there is one interface. Hand it to the EtherCAT master and you have no TCP/IP, no OPC UA server, no remote access — the whole four-port block is now the fieldbus.
There is a second problem: a store-and-forward switch adds latency, and more importantly adds variable latency. Jitter is what deterministic protocols cannot tolerate. Even where a switch is nominally in the path, it has to be one designed for the job.
Bandwidth is shared, not multiplied
Four Gigabit sockets behind one controller do not give you four Gigabits. Everything converges on a single upstream link. For GigE Vision cameras — where the entire reason for multiple ports is to avoid sharing — this defeats the purpose, and the symptom is dropped frames rather than a clean failure.
2. When an internal switch is the right answer
Being fair to the architecture, because this is not a “switch bad” article.
An internal switch is genuinely useful when:
- You are daisy-chaining devices in a line topology. Many machine installations connect device to device rather than back to a central cabinet. An internal switch saves an external one and saves cabinet space.
- You want ring redundancy and the switch is managed and supports MRP or RSTP. Ask specifically — most embedded switches are not.
- Cost matters more than isolation. A switch chip is cheaper than three additional controllers and consumes fewer PCIe lanes.
- You simply need more sockets on one flat network — several sensors, a camera and an HMI all on the same subnet.
The engineering reason switches appear at all: PCIe lanes are a finite budget. Entry-level embedded processors provide a limited number, and four independent Gigabit controllers consume four of them. On a compact board that also needs storage, wireless and expansion, those lanes compete. A switch chip turns one lane into four sockets. That is a legitimate design trade-off — it is just one you should make knowingly rather than discover afterwards.
3. How to verify it yourself in thirty seconds
This is the part worth keeping. You do not need the supplier’s cooperation to find out.
On Linux
# Count the network interfaces the OS actually sees
ip -br link show
# Four independent controllers → four entries (eth0..eth3, or enp1s0, enp2s0, …)
# One controller behind a switch → one entry
# Confirm each interface sits on its own PCI device
lspci | grep -i ethernet
# Show the PCI bus address behind a specific interface
ethtool -i eth0
# bus-info: 0000:01:00.0 ← each independent NIC has a distinct addressFour independent controllers appear as four separate PCI devices with four distinct bus addresses. A switched block shows one.
On Windows
Device Manager → Network adapters. Count the entries. Four independent controllers list four adapters; a switched block lists one.
The unplug test (works on any OS, no tools needed)
Connect all four ports. Unplug one cable and watch link state:
ethtool eth0 | grep "Link detected"- Independent controllers: exactly one interface loses carrier.
- Internal switch: the single interface stays up — the OS cannot see that a port went down at all.
That last behaviour has an operational consequence worth noting: on a switched block, your software cannot detect a disconnected cable on an individual port. For a system that needs to alarm on a lost device connection, that is a functional limitation, not just an architectural one.
Ask for the block diagram
Before purchase, the single most useful request is the block diagram, not the spec sheet. It shows whether each port traces back to its own controller and lane, or converges on a switch. A supplier who has one sends it; a supplier who does not have one has told you something.
4. What to ask a supplier
Four questions, in order of how much they reveal:
- Are the four ports independent controllers or an internal switch? A direct question deserves a direct answer.
- What is the controller part number? Not “Intel” — the part number. Intel i210, i211, i225, i226 and Realtek RTL8111/8125 have materially different capabilities for EtherCAT, hardware timestamping and sustained load. We cover that comparison in our guide to four-LAN DIN rail controllers.
- If there is a switch, is it managed? VLAN tagging, port mirroring and ring protocols are only available on a managed switch.
- Can you send the block diagram? The answer to this one validates the first three.
A note on mixed designs. Some boards combine both — two independent controllers plus a two-port switched block, for example. That is a reasonable design, but it means “4× LAN” describes three usable interfaces, not four. The block diagram is the only reliable way to see it.
5. A worked example
The BI03-IC DIN rail edge controller uses four Intel i210 Gigabit controllers, one per port, each on its own PCIe lane. On Linux this presents as four separate interfaces with four distinct PCI bus addresses, which is what makes independent subnet assignment and EtherCAT master operation possible on it.
That design choice costs four PCIe lanes. It is the right trade-off for a controller intended for network segmentation and fieldbus work, and the wrong one for a device whose job is to fan out a single network to several devices — which is exactly why both architectures exist.
Frequently asked questions
Can I use VLANs instead of independent controllers? Only if the internal switch is managed, which most embedded switches in industrial PCs are not. With independent controllers, VLANs are available at the OS level per interface. With an unmanaged switch, they are not available at all.
Does an internal switch add latency? A store-and-forward switch adds latency, and — more significantly for control applications — adds variability. Deterministic fieldbus protocols are sensitive to jitter rather than to absolute latency.
Can an EtherCAT master run through an internal switch? Not in a normal design. The master requires exclusive control of a network interface and deterministic transmission timing. A switched block presents one interface, and a standard unmanaged switch is not in the timing path the master expects.
How do I know if my application actually needs independent controllers? If any of these apply, it does: separate subnets, OT/IT segmentation, EtherCAT or another master-mode fieldbus, GigE Vision cameras needing dedicated bandwidth, or software that must detect a per-port cable disconnection. If you only need more sockets on one flat network, a switch is simpler and cheaper.
Why would a manufacturer use a switch without saying so? Usually because PCIe lanes are limited on the processor and a switch chip is the only way to provide four sockets. It is a legitimate engineering decision. The problem is only that “4× LAN” on a datasheet does not distinguish between the two.
Working with VstrongTech
If you are specifying a system where port architecture matters, tell us what each port needs to do — which subnet, which protocol, which trust zone — and we will confirm which of our configurations fits, including the block diagram.
If a switched design would serve the application better and cost less, we will say so.