The first time this question bites is usually during a power budget review, and it looks like a simple mismatch. Eight cameras are listed, the switch advertises 370 watts, and the arithmetic seems to work out comfortably. Then two of those cameras turn out to be PTZ models with heaters, the switch starts refusing ports, and the explanation turns out to involve a word that appears on almost every datasheet and is explained on almost none of them: class.
Part of the confusion is structural. Power over Ethernet carried power on the same cable as the data from the beginning, and the 4-pair amendment that raised the ceiling describes itself as increasing maximum device power "by utilizing all four pairs in the specified structured wiring plant," while extending "the power classification information exchanged during negotiation" (IEEE 802.3bt-2018). Two ideas are packed into that sentence: more pairs, and a negotiation. The negotiation is where classes live, and it is the part that decides whether a port will actually turn on.
It is also the part most published explanations get wrong. A survey of the pages currently ranking for this topic found that not one of them explains why a port reserves power before it enables, not one states cable loss as a percentage, and not one works through a multi-port budget example. Two of them mislabel Types as classes, and three give three different wattages for the same standard parameter. So the goal here is narrow and practical: explain what the class handshake actually does, give the complete Class 0 to Class 8 table with both wattage columns, correct the cable loss figure that keeps getting repeated, and show how class decides what an eight or sixteen port unit can really power.
The Short Answer, in One Paragraph
A PoE class is the power level a device requests and a port reserves, and it runs from Class 0 to Class 8. Class 0 through 4 are the older two-pair levels, from 4 W up to 30 W at the port; Class 5 through 8 are the 4-pair levels introduced by 802.3bt, from 45 W up to 90 W at the port. The device always receives less than the port supplies, because the difference covers cable loss. The class is what the port reserves from its shared budget before it turns on.
What a Class Actually Describes
A class is not a property of a cable, a switch, or a standard. It is a number a powered device presents to the port it is plugged into, which tells the port how much power to reserve for it.
Class, Type, and the naming that trips people up
The confusion starts because two numbering systems run in parallel and both get called "PoE class" in casual use.
Type describes the supplying side — the standard revision, the pairs used, and the maximum the port can deliver. There are four types: Type 1 (802.3af), Type 2 (802.3at), Type 3 and Type 4 (both 802.3bt).
Class describes the powered device's request — how much it wants, from Class 0 to Class 8. Each class sits inside a type, and several classes share one type.
The two systems overlap just enough to cause real errors. Class 4 belongs to Type 2, not Type 4. Class 3 is an 802.3af level, so calling it "PoE+" is wrong; PoE+ is the 802.3at level, which is Class 4. And a port can support Type 3 while a device on it asks for a much lower class — the type is a ceiling, the class is a request.
Why the same wattage appears under two different numbers
Every class has two wattage figures, and the gap between them is not an error or a marketing margin. The first is what the port must be capable of supplying. The second is what the device is guaranteed to receive at its own input, after the cable has taken its share.
That is why the same physical class shows up as 15.4 W in one place and 13 W in another. They are the same class measured at two different points.
The Handshake: How a Port Decides What to Deliver
Before any of this matters, the port has to establish that something is on the other end worth powering. The sequence is defined and deliberate, and understanding it explains why a device that "should" work sometimes does not.
Detection comes before power
The port starts by applying a low voltage and measuring. A compliant device presents an effective resistance of about 25 kilohms across the powered pairs, and the port confirms this by measuring at two points and checking that the slope falls inside a narrow acceptance window. During detection the port is not permitted to exceed 30 volts or 5 milliamps, so a device that does not present the right signature cannot be damaged by the attempt.
This is the gate that non-standard "passive" injectors skip entirely, which is why a passive injector can put full voltage onto a port with nothing compatible attached.
Classification, and the signature the device presents
Once detection succeeds, the port applies a classification voltage — in the region of 15.5 to 20.5 volts — and the device responds by drawing a specific current. That current is the signature, and the port reads it to determine the requested class.
| Signature current | Class requested |
|---|---|
| 1 – 4 mA | Class 0 |
| 9 – 12 mA | Class 1 |
| 17 – 20 mA | Class 2 |
| 26 – 30 mA | Class 3 |
| 36 – 44 mA | Class 4 |
Classes above 4 are encoded differently: the device presents the Class 4 signature across a sequence of classification events, and the number of additional events tells a capable port which higher class is being requested. A device asking for Class 8 presents Class 4 first, then the pattern that identifies it as Class 8. A port that only produces one event never sees past the first signature, so it treats a high-power device as a Class 4 request — which is exactly the behaviour that keeps older switches safe when someone plugs in a newer camera.
Why the port reserves power before it turns on
This is the step almost nothing explains, and it is the reason a switch with plenty of advertised wattage can still refuse a port.
A port does not hand out power first and check the budget afterward. During classification it discovers two things at once: how much the device wants, and how much it is allowed to draw. The port then reserves the worst-case power for that class from a shared supply before it applies operating voltage.
The reservation is deliberately pessimistic. For each class, the port must hold back the maximum it might have to source under the worst combination of low supply voltage and maximum cable resistance:
| Class | Power reserved per port |
|---|---|
| 1 | 4.0 W |
| 2 | 6.7 W |
| 3 | 14.0 W |
| 4 | 30.0 W |
| 5 | 45.0 W |
| 6 | 60.0 W |
| 7 | 75.0 W |
| 8 | 90.0 W |
If the sum of reservations across all powered ports would exceed the supply, the port does not negotiate upward. The device is either demoted to a lower class or receives no power at all. A port is never permitted to grant more than the device requested, so demotion always moves down.
The Full Class Table
Here is the complete picture, with both wattage columns and the details that most tables leave out.
| Class | PSE output | PD input | Usable range at the device | Type | Pairs | Introduced by |
|---|---|---|---|---|---|---|
| 0 | 15.4 W | 13.0 W | 0.44 – 12.94 W | Type 1 | 2 | 802.3af |
| 1 | 4.0 W | 3.84 W | 0.44 – 3.84 W | Type 1 | 2 | 802.3af |
| 2 | 7.0 W | 6.49 W | 3.84 – 6.49 W | Type 1 | 2 | 802.3af |
| 3 | 15.4 W | 13.0 W | 6.49 – 12.95 W | Type 1 | 2 | 802.3af |
| 4 | 30.0 W | 25.5 W | 12.95 – 25.50 W | Type 2 | 2 | 802.3at |
| 5 | 45.0 W | 40.0 W | 40 W | Type 3 | 4 | 802.3bt |
| 6 | 60.0 W | 51.0 W | 51 W | Type 3 | 4 | 802.3bt |
| 7 | 75.0 W | 62.0 W | 62 W | Type 4 | 4 | 802.3bt |
| 8 | 90.0 W | 71.3 W | 71.3 W | Type 4 | 4 | 802.3bt |
Reading the two wattage columns
The PSE output column is what the supplying port must be capable of. The PD input column is the floor the device can rely on at its own connector. The usable range is where the class actually sits between its neighbours — a device drawing 4 W belongs in Class 2, not Class 1, because Class 1 stops at 3.84 W.
Note that Classes 0 and 3 share identical wattages. That is not a typo. Class 0 was defined as the fallback for devices that implemented no classification circuit at all, and it was assigned the Class 3 power level so an unclassified device would still work.
Where Class 0 sits in a modern system
Class 0 is effectively a legacy behaviour rather than a design choice. It is defined in the original clause but deprecated in the 4-pair clause, and devices built to the newer standard are required to support classification, so Class 0 does not exist for them. When a newer port meets an older unclassified device, it assigns it Class 3 and moves on.
Because the two share a power level, this is harmless in practice. It becomes relevant only when you are auditing why a port shows a particular reservation and the device's documentation never mentioned a class at all.
Cable Loss: What Actually Disappears Over 100 Meters
The gap between the two wattage columns is the part of the story that circulates most inaccurately, and it is worth getting right because it changes how much headroom you plan for.
The gap between the two columns
Expressed as a share of what the port supplies:
| Type | Port output | Device input | Difference | Loss as a share of port output |
|---|---|---|---|---|
| Type 1 | 15.4 W | 13.0 W | 2.4 W | 15.6% |
| Type 2 | 30.0 W | 25.5 W | 4.5 W | 15.0% |
| Type 3 | 60.0 W | 51.0 W | 9.0 W | 15.0% |
| Type 4 | 90.0 W | 71.3 W | 18.7 W | 20.8% |
The share grows at the top of the range because the higher classes push more current through the same channel, and loss scales with the square of current.
Why the commonly quoted loss figure is wrong
A figure of 25 percent gets repeated often enough that it reads as authoritative. It is not what the standard specifies. The Ethernet Alliance's technical team, writing specifically to correct this, puts the real position plainly: at the interoperability boundary conditions the standard supports, "relative cable losses of 15 percent seem to be the norm," and operation at 90 W "even sees a potential cable loss of 20 percent" (Ethernet Alliance on PoE cable losses).
So the numbers to carry in your head are 15 percent as the normal worst case and roughly 20 percent at the very top of the range. The 25 percent figure appears to come from conflating cable loss with a completely different loss — the efficiency of the device's own internal power converter, which does shed somewhere between 10 and 25 percent of what arrives, but as heat inside the device rather than in the cable.
There is also a 25 percent figure in the standard's own documentation, and it means something else again: the saving that an optional feature called Autoclass delivers by measuring actual cable resistance instead of assuming the worst case. On a Class 8 port that optimisation reduces the allocation from 90 W to about 67.8 W — a saving of roughly a quarter, and the reason a well-implemented switch can power more high-class ports from the same supply.
What the standards actually limit
Rather than a percentage, the standard limits the channel itself. The worst-case pairset DC loop resistance is fixed at 12.5 ohms, and for 4-pair powering the two pairsets run in parallel, giving an effective 6.25 ohms. A real installed cable almost always measures far better than the limit, which is why measured losses in practice are much lower than the worst case — an Ethernet Alliance study of a large lighting deployment found aggregated losses of about 2 percent, and a smaller system measured around 0.5 percent.
The practical conclusion is that the two-column gap describes a boundary condition, not a typical one. Plan against it, but do not expect your cable to consume it.
The Power Budget Across Eight and Sixteen Ports
The class table becomes useful the moment you put a fixed supply behind it. This is the arithmetic no ranking page currently does.
Working the arithmetic for eight ports
Take a switch with a 370 W supply and a full complement of eight ports. The maximum port count at each class follows directly from the reservation values:
| Class | Reserved per port | Ports a 370 W supply can support |
|---|---|---|
| 3 | 14.0 W | all 8 |
| 4 | 30.0 W | all 8 (12 possible) |
| 5 | 45.0 W | 8 |
| 6 | 60.0 W | 6 |
| 7 | 75.0 W | 4 |
| 8 | 90.0 W | 4 |
The lower classes never trouble a supply this size. The interesting zone starts at Class 6, where the reservation is high enough that the supply, not the port count, becomes the limit.
Working the arithmetic for sixteen ports
The same supply scaled up to 740 W across sixteen ports:
| Class | Reserved per port | Ports a 740 W supply can support |
|---|---|---|
| 4 | 30.0 W | all 16 (24 possible) |
| 5 | 45.0 W | 16 |
| 6 | 60.0 W | 12 |
| 7 | 75.0 W | 9 |
| 8 | 90.0 W | 8 |
The pattern to notice is that a sixteen-port unit is rarely limited by its port count at the top of the range. At Class 8, a 740 W supply powers eight ports — half the chassis. Buying a sixteen-port switch for sixteen Class 8 devices without checking the budget means buying a chassis that will refuse eight of them.
Mixed classes are where budgets actually break
Real installations are never uniform, and the reservation is per port rather than averaged. Consider the eight-port example again at 370 W: four ports at Class 4 reserve 120 W, and four ports at Class 6 reserve 240 W. The total is 360 W against a 370 W supply — it fits, with ten watts to spare.
Now add a fifth Class 4 device. That port needs 30 W, taking the total to 390 W, which the supply cannot cover. The new device is not refused outright; it is demoted to a lower class, which means it may boot and then behave unpredictably under load rather than failing cleanly.
What happens when the budget will not stretch
Demotion is the behaviour to expect, and it deserves more attention than it gets because the symptom is misleading. A demoted device typically powers up, connects, and works — until it asks for more than its reduced allocation allows. On a camera with a heater or a PTZ motor, that shows up as intermittent reboots in cold weather or when the motor engages, which looks like a faulty camera rather than a power budget problem.
Autoclass exists precisely to soften this. By measuring the actual resistance of the installed cable instead of assuming the worst case, a capable switch can allocate closer to what the device genuinely needs and squeeze more ports out of the same supply.
Which Class You Actually Need
Class is a property of the device, so the practical question is what kinds of device sit in which band. What follows is a configuration guide by device type rather than a product list.
Cameras that want the smallest class
A fixed dome or bullet camera recording at 4K draws well under 9 W in typical operation, which puts it comfortably inside the older two-pair classes. The practical consequence is that a full complement of ordinary cameras will almost never strain a switch budget, and the port count — not the wattage — is the limit you hit first. If every device on the switch is a fixed camera, buy ports.
Cameras with heaters, motors, or floodlights
This is where class stops being academic. A PTZ camera with pan and tilt motors draws materially more when the motors engage than when parked, and a camera with an internal heater draws more in cold weather than the datasheet's idle figure suggests. Illuminator arrays and motorised zoom add further peaks. These devices live in the upper classes, and they are the reason to check the reservation table rather than the average draw.
The specific trap is a device that idles low and peaks high. Class allocation is based on the reserved worst case, not the idle figure, so a camera rated at 8 W idle may still request a substantially higher class.
Access points, displays, and other devices on the same switch
If the switch also powers wireless access points, signage, or door controllers, the budget arithmetic changes because those devices reserve their own classes. The mixed-class example above is the common real-world case: a few high-class devices alongside many low-class ones, comfortably inside the supply until one more high-class device arrives. Keep a reservation table for the whole switch rather than a per-device calculation.
The reservation arithmetic only has to be done by hand when devices from different sources share one switch.
When to buy headroom rather than enough
Headroom earns its cost in two situations: when a device on the port might be replaced with a higher-draw model later, and when the switch will run near its budget for long periods. In both cases the question to answer before purchase is not the total wattage on the box but the reservation sum at the highest class you expect to support simultaneously.
A Quick Reference Table
| If the device is… | Expect a class of… | Port must supply | Device receives |
|---|---|---|---|
| A basic fixed camera or sensor | Class 1 – 2 | 4 – 7 W | 3.84 – 6.49 W |
| A fixed 4K camera | Class 3 | 15.4 W | 13.0 W |
| A wireless access point | Class 3 – 4 | 15.4 – 30 W | 13.0 – 25.5 W |
| A PTZ with motors | Class 4 – 5 | 30 – 45 W | 25.5 – 40 W |
| A heater-equipped outdoor camera | Class 5 – 6 | 45 – 60 W | 40 – 51 W |
| A multi-sensor or dual-lens camera | Class 6 – 7 | 60 – 75 W | 51 – 62 W |
| A large dome with illuminators and heater | Class 7 – 8 | 75 – 90 W | 62 – 71.3 W |
These are indicative bands based on typical draw, not guarantees. The definitive figure is the class the device presents when it is classified — which is what the port will reserve, regardless of what the datasheet's average suggests.
What a Complete PoE System Settles for You
Everything above is engineering work, and a PoE security camera system sold as a complete unit has already done it — which means you do not have to work through the class table, the reservation values, or the loss figures to end up with a system that runs correctly. The recorder's built-in ports and their power budget were sized against the cameras in the box, and the classes were matched to what those cameras actually draw. Each camera then runs on one cable carrying both video and power: no separate supply, no outlet at the mounting point, no injector when a camera sits a long way from a socket, and no monthly fee when footage is stored locally on the recorder's own drive. Expansion becomes a question of free ports and free channels rather than a re-planning of the supply, and one interface manages every camera at once. If your camera count and resolution are settled, buy a complete system such as the 8 or 16 channel PoE range from 4COVR and skip the power design entirely — confirming only that the published budget covers the cameras in the box. Keep the arithmetic above for the cases where you assemble a system from separate parts, or add another source's devices to an existing switch.
Questions People Ask Before They Buy
What is the difference between a PoE class and a PoE type?
A type describes the supplying side: the standard revision, the pairs used, and the maximum the port can deliver. There are four types, from Type 1 to Type 4. A class describes the device's request, from Class 0 to Class 8. Every class belongs to a type, and several classes share one type.
Is Class 3 the same as PoE+?
No, and this is one of the most common errors in published explanations. Class 3 is an 802.3af power level. PoE+ is 802.3at, which corresponds to Class 4. The two differ by roughly a factor of two in both port output and device input.
What happens if a device needs more power than its class allows?
The port reserves according to the class the device presents. If a device draws more than its allocation after start-up, the port will disconnect it. A device that draws less than its class allows is fine — the reservation is a ceiling, and unused headroom is simply held rather than delivered.
Can a Class 8 device run on a switch that only supports Class 4?
It will power on, but not at full capability. A port that produces only the first classification event cannot see past the initial signature, so a high-power device is treated as a Class 4 request and allocated 30 W. Whether the device functions depends on whether it can operate within that reduced budget, which for a camera with heaters or motors is often marginal rather than absolute.
Do I need a different cable for the higher classes?
The cable requirements tighten at the top of the range rather than changing category. The standard fixes a worst-case pairset loop resistance of 12.5 ohms, halving to an effective 6.25 ohms when four pairs run in parallel, and the higher classes push more current through that channel. Solid copper cable of a standard category meets these figures; copper-clad aluminium, which has substantially higher resistance, does not.
What does Class 0 mean?
Class 0 is the fallback for a device that implements no classification circuit and presents no power request. It shares the Class 3 power level, so an unclassified device still receives enough to operate. Modern devices built to the 4-pair standard are required to support classification, so Class 0 is effectively a legacy behaviour.
Where This Leaves the Decision
Class is a request, and the port's job is to reserve for it before anything is switched on. That single mechanic explains nearly every confusing behaviour in this area: why a switch with a large advertised wattage refuses a port, why a camera reboots intermittently in cold weather, and why a device that ought to fit does not.
Three numbers are worth carrying forward. The two-column gap is about 15 percent in normal worst-case conditions and roughly 20 percent at the top of the range, not the 25 percent that circulates. The reservation per port is what the budget sees, not the device's average draw — so the sum at the highest class you expect to support simultaneously is the figure that decides whether the chassis is adequate. And a high-power device on a lower-type port will usually still run, quietly demoted, which is the failure mode least likely to be diagnosed correctly.
If you are specifying a switch, start from the reservation table and work backwards to the port count and the supply. If you are buying a kit, the sizing has usually been done for you, and the thing worth confirming is that the recorder's published budget covers the cameras in the box at their classification rather than their idle draw.