Type 1 sources 15.4 W at the port and Type 4 sources 90 W. But a Type 4 device is only guaranteed 71.3 W, because 18.7 W is burned in the copper. Every famous PoE number is Ohm's law applied twice, and a hot cable bundle takes some of them back.
Four ideas to carry into the next specification conversation you have.
802.3af, 802.3at and 802.3bt are all superseded amendments. What exists is IEEE Std 802.3-2022: Clause 33 for Types 1 and 2, Clause 145 for Types 3 and 4. And the standard's own text never uses the words PoE, PoE+, PoE++ or UPOE.
PSE is the source, PD the device, and the gap between them is one equation: PPD = PPSE − (PPSE/V)² × R. Type 1 gives 12.95 W, Type 2 25.5 W, Type 3 51 W and Type 4 71.3 W. Four numbers everyone memorises, one equation nobody shows them.
Four pairs halve the resistance, and then Class 8 spends all of it and more on current: 865 mA per pair against 600 mA. So Type 4 is the least efficient PoE, not the most: 79.2% worst case against 85.0% at Class 4, because loss goes as current squared.
IEEE defines two axes and neither is a marketing name: Type is a property of the equipment, Class is the result of a negotiation on that port and the thing that changes at run time. Specify the Class and the guaranteed watts at the device, not the Type and the port watts.
Computed from the published equations, not copied from a table. Channel at the standard's own worst case: 20 Ω two-pair for Type 1, 12.5 Ω for Type 2, 6.25 Ω four-pair for Types 3 and 4.
| PoE Type | PSE / PD power | Typical loads |
|---|---|---|
| Live standard | IEEE Std 802.3-2022 | 802.3af/at/bt are superseded amendments. Types 1 and 2 live in Clause 33, Types 3 and 4 in Clause 145 |
| Type 1, Class 3 | 15.4 W / 13.0 W | The equation gives 12.95 W, but 802.3at rounded it to 13.0 W in 2009. "12.95" is a stale citation |
| Type 2, Class 4 | 30 W / 25.5 W | Two-pair, 12.5 Ω pairset. A Type 3 PSE may also stop here: four pairs are mandatory only from Class 5 |
| Type 3, Class 6 | 60 W / 51 W | Four-pair, 6.25 Ω. "Type 3" alone guarantees only Class 1, 3.84 W at the device |
| Type 4, Class 8 | 90 W / 71.3 W | Optional: a Type 4 PSE need only support Class 7, which is 75 W at the port and 62 W at the device |
| Minimum port voltage | 44 / 50 / 50 / 52 V | Type 1 to Type 4. A Type 4 clause that says 50 V accepts a non-conforming switch |
| Channel budget | 6.25 Ω four-pair | The standard budgets resistance, never metres. 12.5 Ω two-pair, from ISO/IEC 11801 Class D |
| Worst-case efficiency | 79.2% at Class 8 | Against 85.0% at Class 4 and Class 6. Type 4 is the least efficient PoE, because loss goes as current squared |
| Bundle limit at 45 °C | 865 mA per pair | TIA TSB-184-A model: Cat 6A, 100-cable bundle, every pair on. Cat 6 reaches it at 42.7 °C and Cat 5e at 39.2 °C |
| Saudi code | SBC 401 ch. 716 | New in the 2024 edition, under the section headings of IEC 60364-7-716:2023, which caps any conductor at 750 mA |
Five practical gotchas that catch new BAS designers.
IEEE 802.3 says nothing at all about a switch's total budget: it is written for a single PSE, one link segment and a single PD. So ask the question that breaks every quotation, how many ports can run at 90 W simultaneously? One major vendor answers 21 in its own datasheet, on both the 24-port and the 48-port model. The budget belongs to the power supplies, and it moves with mains voltage, where Saudi's 230 V gets the better number.
TIA TSB-184-A models a 100-cable bundle with every pair energised: I = ρ × 1000 × √(1 − t/60) mA per pair. For Cat 6A at 45 °C it gives 865 mA, exactly what a 90 W Type 4 port draws per pair at 52 V, and the same model reaches that edge at 42.7 °C for Cat 6 and 39.2 °C for Cat 5e. One manufacturer's own table for its Cat 6A says 960 mA in free air but 843 mA in conduit, and 1,632 mA once the bundle is cut to 24 cables. Two models, two answers: state the bundle size, the pathway and a design ambient, because nothing in the switch tells you.
The 25 kΩ signature only says "I am a powered device". Then the switch holds 15.5 to 20.5 V and the device asks with the current it draws: a Class 8 device answers "Class 4" twice, so an older switch sees a harmless load, and changes its answer from the third event. The switch answers with the number of events: one grants Class 3, two or three Class 4, four Class 6, five Class 8. A switch short of budget sends fewer, so a 90 W luminaire can come up at 51 W or 25.5 W without any fault, and it must hold its draw inside the Class it was given.
A midspan injector adds PoE without replacing the switch, which is why retrofit projects use them. Two cautions worth writing into the specification: the injector's cords count in the channel resistance budget, and unplugging a live PoE port is officially a misuse of the connector. IEC 60512-99-002 says so in as many words, qualifies it over 100 unmate cycles rather than the RJ45's 2,500 mechanical ones, and explicitly tolerates visible contact damage in its pass criteria.
PNNL, for the US Department of Energy, bench-tested a PoE lighting system built to ANSI C137.3. With the switch and 49 m cables included it drew 13 to 19 percent more than the luminaires' own input power, because PoE switches take AC in and convert it: the conversion moves, it does not disappear. For PoE to save energy, a PoE luminaire has to beat its AC twin by more than that overhead, and nobody has published that measurement. Trace the famous 10 to 20 percent saving and it ends at a 2018 simulation of whole-building DC distribution with PV and batteries, which is not a PoE study at all. Most "PoE luminaires" are also not PoE devices: the Ethernet stops at a node in the ceiling. Specify PoE for the control granularity and for moving the AC front end out of a hot ceiling void, both of which are real.
Six lines, each naming a document, a quantity or a unit. The deck's slide 16 carries the same six, from the same source file.
1. PSE: IEEE Std 802.3-2022 Clause 145, Type 4, four-pair, minimum port voltage 52 V.
2. Contractor guarantees [X] W at the PD power interface, not at the port, calculated with the measured channel resistance.
3. Measured pairset DC loop resistance recorded per channel, not above 12.5 Ω per pairset.
4. Design ambient in the cable pathway stated by the designer, never assumed at 20 °C.
5. Bundle size stated; temperature rise calculated to TIA TSB-184-A at that ambient with all four pairs energised, within the cable's 60 °C rating.
6. Switch PoE budget stated for the supplied power supplies at project mains voltage, with the number of ports supportable at the specified Class at once.
Line 1 is the one most often written wrong: 50 V is the Type 2 and Type 3 floor. A Type 4 port may not run below 52 V.
We size PoE switches against the guaranteed power at the device rather than the badge on the port, state how many ports can run at the specified Class at once, and check the cable bundle thermally against TIA TSB-184-A at a stated design ambient rather than a 20 °C one.
Practical lighting know-how from NLC Academy. No spam, unsubscribe anytime.
By subscribing you agree to receive NLC emails.
The standards and measurements behind every number on this page.
An 18-slide explainer, ready to post straight to LinkedIn as an image carousel.
