Sixty watts is the safe number. Ninety watts is where four-pair PoE stops being a bigger version of PoE+ and turns into a different electrical problem. That is the short version, and it is the reverse of what most people assume when they swap an 802.3at switch for an 802.3bt one.
I spent last Tuesday in a ceiling void above a packing line, laptop balanced on a ladder rung, because eight access points had been swapped for Wi-Fi 7 units and two kept dropping between 13:00 and 15:00. The switch had powered a 60 W AP on that port for six months without complaint. The new unit asked for Class 8 instead of Class 6, and the cable had opinions about that. Not a broken link — a link that was fine at 60 W and marginal at 90 W, a distinction the port log does not make for you.
Four Pairs Is Not Four Times the Headroom
802.3bt's defining move is easy to misread: the PSE energizes both pairsets at once instead of one. Two-pair PoE puts positive on one pair and negative on another, leaving the remaining two pairs idle. Four-pair mode uses all eight conductors — two pairs positive, two pairs negative — so the port can move roughly twice the power without asking any single conductor to carry twice the current.
That last clause is the part people skip. Here is the whole electrical picture in one table. The first five columns are standard; the last one is my own division, and it is the column that decides whether your cable is comfortable.
| Standard | PSE output | PD receives | Pairs powered | Max current per pairset | Current per conductor at that limit |
|---|---|---|---|---|---|
| 802.3af (Type 1) | 15.4 W | 12.95 W | 2 | 350 mA | 175 mA |
| 802.3at (Type 2) | 30 W | 25.5 W | 2 | 600 mA | 300 mA |
| 802.3bt Type 3 | 60 W | 51 W | 4 | 600 mA | 300 mA |
| 802.3bt Type 4 | 90 W | 71.3 W | 4 | 866 mA (960 mA peak) | 433 mA |
Read the bottom two rows against the one above them. 802.3at pushes 600 mA through a pairset, which is 300 mA per conductor. Type 3 pushes the same 600 mA through a pairset — the standard never raised that limit. What changed is that two pairsets now work in parallel, so total current climbs to about 1.2 A while each conductor still sees 300 mA. The port doubled its output and no individual wire got hotter. Type 4 is where that stops being true: 866 mA per pairset is 433 mA per conductor, and heat follows the square, so (433 ÷ 300)² is just over 2. Every conductor in the cable now dissipates roughly twice what it did at 30 W.
Which is why I call 60 W the least scary row in that table. It looks like a big jump because the number doubled. Electrically it is the same demand on the same copper, spread over twice as much of it.
The Wattage Check That Tells You Which Row You're Actually In
You can reconstruct all four rows from the wattage differences alone, which is useful when a datasheet hands you numbers that do not obviously agree. The gap between PSE output and PD input is the channel allowance, and on a 100 m channel it has to be I²R.
- 802.3at: 30 − 25.5 = 4.5 W of allowance. At 600 mA that implies 4.5 ÷ 0.6² = 12.5 Ω, exactly the Category 5 loop figure the standard is built around.
- Type 3: 60 − 51 = 9 W. Two pairsets at 600 mA each: 2 × (0.6² × 12.5) = 9.0 W. Same resistance, twice the paths.
- Type 4: 90 W at the 52 V minimum PSE voltage is 1.73 A total. Across two pairsets, 866 mA each. Then 2 × (0.866² × 12.5) = 18.7 W, and 90 − 18.7 = 71.3 W at the PD — matching the standard's own figure, which is the check that the model is right rather than arithmetic that merely looks plausible.
- Per conductor at Type 4: 866 ÷ 2 = 433 mA, which is 1.44× the 300 mA on a PoE+ conductor. Squared, 2.08× the heat per wire.
If your numbers miss 25.5, 51 and 71.3, you are using a different channel resistance — often because you measured a short patch cord and generalised it to a 90 m run. The 12.5 Ω is a limit for the whole channel, not a property of the cable on your bench.
One caveat: 866 mA is the nominal split, the case where the two positive pairs share perfectly evenly. They do not. Which brings us to what actually breaks PoE switches at 90 W.
Where 90 W Really Bites: Pair-to-Pair Unbalance
Two pairs of the same polarity are wired in parallel, and parallel conductors only share in proportion to their conductance. Copper tolerance, contact resistance, how well the IDC blocks were punched, a patch cord swapped for a thinner one at the desk — all of it shifts the split. The IEEE working group calls the per-pair figures nominal for exactly this reason, and the Ethernet Alliance's overview quantifies how far one pair can drift.
| Class | PSE output | Current on a pairset, perfect balance | Worst pairset in the same class | Extra heat on that pairset |
|---|---|---|---|---|
| Class 5 | 45 W | 450 mA | 550 mA | +49% |
| Class 6 | 60 W | 600 mA | 682 mA | +29% |
| Class 7 | 75 W | 721 mA | 784 mA | +18% |
| Class 8 | 90 W | 866 mA | 938 mA | +17% |
Note the shape of that column. The percentage penalty shrinks as the class rises; the absolute current does not. 938 mA on one pairset is 469 mA per conductor, and on a 100 m run that is where the temperature margin you thought you had quietly walks away. A link built for 60 W with 8% unbalance is fine. The same link asked for 90 W is running one pairset 17% over its nominal share, on a channel whose resistance was already at the limit.
I got this wrong the first time I met it. The APs were rebooting, the switch showed both ports as up, and my first two hours went into the access points — firmware, channel plan, a suspicion about a buggy PoE driver. They were fine. What pointed at the cable was a log line showing the port renegotiating down a class and climbing back. A PD that browns out and recovers looks exactly like a device with a software fault, and the recovery is fast enough that a ping test never catches it.
Certifiers report resistance unbalance within a pair, because TIA-1152 puts a limit on it. Pair-to-pair balance across the two positive pairs is not something a field test hands you as pass or fail. You measure it: disconnect both ends, use a four-wire ohmmeter or a tester's DC resistance mode, compare the pairs that carry the same polarity. A 5% spread on a 90 W port is worth a conversation; 10% is worth replacing the run.
Single-Signature and Dual-Signature Decide What the Port Allocates
802.3bt introduced a distinction with nothing to do with copper and everything to do with whether your budget survives contact with reality. A single-signature PD presents one detection signature across both pairsets and negotiates one class. A dual-signature PD presents two independent signatures and can be allocated two classes — in effect two 25.5 W loads on one jack.
On a switch with a fixed per-port ceiling the consequence is immediate. A dual-signature device on a Type 3 port may come up as two Class 4 slots: 51 W of reservation for something that draws 38 W. The single-signature version of the same hardware reserves one Class 6 slot. Same device on paper, different arithmetic in the switch. I have watched a 24-port switch refuse four ports at commissioning purely because the APs were dual-signature and the reservation model added them in pairs.
There is a second-order effect. If a dual-signature PD loses one pairset, the other half keeps running — the device degrades instead of dying. That sounds like a feature until you are troubleshooting a camera that works but has no heater and the port still reads "delivering power". Single-signature devices are simpler to reason about, and on a mixed-vendor network I prefer them. Either way, know which one is on the end of the cable before you size the budget.
Autoclass complicates it in a good way. A PSE that supports it measures actual draw over the first few seconds and releases the unused reservation — the difference between a 60 W port costing 60 W of budget and costing 41 W. Not every switch implements it, and fewer document it.
What I Check Before I Trust a 90 W Port
This is the sequence I run before signing anything off at Type 3 or Type 4. Twenty minutes for a dozen ports, and it has saved me at least three return visits.
- Confirm the class that was actually allocated, not the class you asked for. The switch's per-port page is the only place this is visible. A port that negotiated Class 4 on a 90 W device will work, badly.
- Measure DC resistance on both pairs of each polarity. Four readings per link, compared as pairs. Anything over 5% deserves a second look, and over 10% means the run comes out.
- Read the loaded voltage at the far end, not at the switch. A 90 W port sagging below 50 V under load is telling you about the channel, and the PD will confirm it later at a worse time.
- Check the bundle before you check the ambient. Six 90 W cables in a 25 mm conduit behave nothing like one. This is where cable category and bundle fill stop being procurement questions and become temperature questions.
- Look at the patch cords. Cheapest part of the channel, most likely to be 26 AWG stranded. A 90 W port with a 26 AWG cord at each end has two unnecessary resistors in series.
- Leave the load running an hour and read port temperature, if the platform exposes it. Most do not, and a cheap IR thermometer on the jack face beats nothing.
For 60 W work I skip steps 2 and 6. The conductor current matches PoE+, so a link that already ran a 30 W device will not change its thermal life — that is not a guess, it is the table at the top of this page. Sites that get into trouble at 60 W were marginal at 30 W already: CCA cable, a 90 m run, three couplers in the ceiling. The upgrade did not create the problem. It moved the failure from occasionally to every afternoon.
Questions I Get Asked About Four-Pair Powering
Does 802.3bt require Cat6A?
No. Category 5e is the standard's minimum for Type 3 and Type 4. What changes is thermal planning: TIA TSB-184-A and IEC TS 29125 assume you derate bundle ambient when all four pairs carry current at the limit, and at 90 W in a dense bundle that derating is what pushes projects to Cat6A. The electrical requirement did not move; the margin did.
Can I run 90 W over the Cat5e already in the ceiling?
Sometimes. It depends on three things the cable label will not tell you: bundle size, ambient, pair-to-pair balance. One 90 W run in free air over 30 m is usually fine. Twelve in a shared conduit above a server room is a different project. Do not take "it powered up" as evidence.
Why does a 60 W access point show 51 W at the port?
Because 51 W is the guaranteed minimum at the PD for that class, and the 9 W gap is the channel allowance: two pairsets at 600 mA across 12.5 Ω. The PD is not being shortchanged — the switch reserved 60 W, the cable ate 9 W, and that is the deal the standard struck so a 100 m run could still deliver a defined number.
What is the difference between Type 3 and Class 6?
Type is the PSE's capability and the physical layer — pairs, voltage band, current limit. Class is what a specific PD negotiates inside that. As the class tables show, a Type 3 port serves Classes 0 through 6, and Class 6 draws 51 W. Classes 7 and 8 need a Type 4 port, so "supports 802.3bt" is not the same as "supports 90 W".
Do I need special RJ45 jacks for 90 W?
Less than you would guess. Contact current is 433 mA per conductor at Type 4 against 300 mA at PoE+, both well inside a normal 1 A rating. The parts that struggle are the magnetics and the PCB: center-tap windings sized for PoE+ saturate, thin traces run hot for eight hours. On the cable side, termination quality matters more than the jack — which is where a poorly punched connector does its damage, quietly, at 80 m.
Will a four-pair PD work on a two-pair PSE?
Only if it was designed to fall back, and many are not. A Type 3 or Type 4 PD that requires four-pair power fails detection on an 802.3at port, which looks identical to a dead device. Check fallback behaviour before you buy — the failure mode is silent, and I have watched a crew replace three cameras that were never broken.
Wrapping Up, With the Numbers Still on the Table
Four-pair powering is good engineering that got marketed as "more watts", and the marketing causes most of the arguments. The watts are not the interesting part; conductors are. At 60 W each one still runs at 300 mA, so the cable work you did for PoE+ stands. At 90 W they go to 433 mA, and the imbalance lands on whichever pairset has the lower resistance.
If you are measuring something that does not match this arithmetic — a link that passes every test and still drops at two in the afternoon — I would like to see the numbers. Send the loaded voltage, the class that was actually allocated, and the four DC resistance readings. Nine times out of ten the answer is in that last set. The tenth time it is the magnetics, and I will happily be wrong about which one it was.