A camera datasheet says the unit needs 12.9 watts. The switch port says it supplies 15.4. That looks like two and a half watts of headroom, so the installer runs the cable, clips the RJ45, and the camera either refuses to boot or comes up, streams for a day, and then starts cycling. Nothing in the numbers explains it, because the two numbers are not measured in the same place.
The 15.4 figure is a power-sourcing-equipment number. It describes what leaves the switch port. The 12.95 figure is a powered-device number. It describes what the device is entitled to find at the far end of the cable. Every reading problem in a PoE install starts with a document that prints one of those figures without saying which side of the cable it belongs to — and at least one major vendor prints 15.4 on the wrong side.
This piece follows the single figure 15.4 through five manufacturer documents, shows which side of the cable it lives on, and works out why a 60-watt switch budget stops at three Class 3 ports.
The Same 15.4 Watts Appears on Both Sides of the Cable
Start with the cleanest phrasing. The Ubiquiti Help Center article "Intro to Networking - Power Over Ethernet (PoE)" describes the first Active PoE standard as an “IEEE standard that supports low power devices and provides a maximum of 15.4W at the PSE.” The three words that matter are the last three. The same article puts the next tier at “IEEE standard that provides a maximum of 30W at the PSE” and the tier above that at “IEEE standard that provides a maximum of 60W (type 3) or 100W (type 4) at the PSE.” Four figures, one qualifier, applied consistently.
The Omada Networks support document "Why my PoE powered device cannot work properly when connected to the PoE Switch?" says the same thing in the language of ports rather than roles, and its sentence carries both sides before it ends: “IEEE802.3af standard defines the maximum output power from a single port to be 15.4W, and it can supply power to the devices under 12.95W (such as IP camera) through Ethernet cable.” Output power from a port, 15.4; devices supplied, under 12.95. One sentence, both columns.
Then the phrasing slips. Cisco's "Configuring Power over Ethernet" chapter for the Catalyst 9400 Series writes that “The PoE+ standard increases the maximum power that can be drawn by a powered device from 15.4 W per port to 30 W per port.” That sentence puts 15.4 on the device side, and so does the table beneath it: the column carrying 15.4 W for Class 0 and Class 3 is headed “Maximum Power Level Required from the Device.” An installer who reads only this page will size a device against 15.4 and be wrong by 2.45 watts.
The resolution is in another Cisco document for a different platform. The "Power over Ethernet" configuration chapter for the Catalyst 6500 Series prints both sides of each class on one line. Its default classification row reads “Class 0: Up to 15.4 W (0.44–12.95 W at the PD; default classification)” and its top 802.3af row reads “Class 3: Up to 15.4 W (6.49–12.95 W at the PD).” The PoE+ row continues the pattern with the qualifier spelled out on both sides: “Class 4: 30.00 W at the PSE (12.95 W to 25.50 W at the PD).”
That settles it. Within one manufacturer's own documentation set, 15.4 is the port-side figure for Class 0 and Class 3, and 12.95 is the ceiling such a device may rely on. The Catalyst 9400 page is the outlier, and the page a search engine is most likely to surface, because it puts 15.4 and 30 in one readable line.
NETGEAR Support's article "Power over Ethernet (PoE) standards and compatibility" keeps the two sides in separate table columns, which is why its figures are the least ambiguous of the five. Its Class 0 row gives a power range of 0.44W-12.95W with a minimum output at the switch port of 15.4W and a maximum output at the switch port of 16.2W. Class 4 gets a power range of 12.95W-25.5W against switch port outputs of 30.0W and 31.6W. Same numbers, columns labelled.
The 2.45 Watts Between the Port and the Plug
Subtract: 15.4 minus 12.95 is 2.45. That subtraction is not printed in any of the five documents, but the reason for the gap is. NETGEAR's article ends its class table with one sentence of explanation: “These standard power ranges are calculated with the maximum cable length of 328 feet (100 meters).”
So the published ranges already assume the worst case the cabling rules permit, which is how the gap is usually read. The Omada document states the same limit from the installation side — the length of the Ethernet cable cannot be over 100m, using Cat.5 or above Ethernet cable. Why the device-side figure lands at 12.95 is not stated on any page collected.
Two consequences follow, and they point in opposite directions from what installers usually assume.
First, a short run does not raise the device's ceiling. The class figures are fixed by classification, not by the length of your patch cable. A Class 3 device three metres from the switch is still a Class 3 device with a 12.95-watt entitlement. The margin exists for the worst case; it is not a credit you get back on a good run.
Second, the same arithmetic scales up. Class 4 is 30.00 W at the PSE against a 25.50 W ceiling at the device, a gap of 4.5 watts. The gap grows with the class. None of the five documents prints a loop-resistance figure or a derivation, so this piece offers no mechanism for that growth.
What the Switch Reserves Is Not What the Device Draws
The second place the 15.4 figure does real damage is the power budget. A switch advertises a total, the installer divides the total by the device draw, and the count comes out too high.
The Catalyst 6500 chapter states the mechanism plainly: “Each PD requires power to be allocated from the chassis power budget.” Allocation, not measurement. The same chapter gives the default size of that allocation: “If no maximum power level is configured, the default maximum power is 15400 milliwatts.” It also gives the range an administrator may set instead: “The configurable range of maximum power using the max keyword is 4000 to 16800 milliwatts.”
Read those three sentences together and the budget arithmetic changes. The unit of account is port-side milliwatts. A camera that actually draws six watts still has 15400 milliwatts reserved against the chassis budget until something tells the switch otherwise. The upper end of that configurable range matches the Catalyst 6500 line “Maximum 16.80 W at the PSE” — about 1.4 watts of configurable headroom above the default.
Now take a small switch with a 60-watt PoE budget and four Class 3 cameras. Three reservations at 15.4 watts come to 46.2 watts, which fits. Four come to 61.6 watts, which does not. The fourth port is refused even though the four cameras together would draw well under 60 watts in service. The Omada document describes the same condition from the symptom side, as the total output power of the PSE exceeding or approaching the maximum output power.
This is the same trap as reading a charger's headline wattage as a promise to a specific device, which the piece on why a 45-watt label is a ceiling and a 9-volt device stops at 27 watts works through for USB-C. In both cases the printed number describes the supply's capability, and the device's own negotiated figure is lower and is the one that governs.
Reading a Device Label Against the Class Table
The practical test is one comparison. Find the figure on the powered device's label or datasheet, then decide which column of the class table it belongs in.
A device label reading 12.9 watts or anything just under 12.95 is a Class 3 device described from the device side, and an 802.3af port supplies it correctly. A label reading 15.4 watts is quoting the port column. No class row above gives a Class 3 device a 15.4-watt entitlement, so a device genuinely needing that much belongs on a Class 4 port with its 25.50-watt device-side ceiling.
The lower classes are less often misread because the two columns are further apart. The Catalyst 6500 class table gives Class 1 as up to 4 W and Class 2 as up to 7 W on the port side; the NETGEAR table gives the matching device-side ranges as 0.44W-3.84W and 3.84W-6.49W, with port outputs of 4.0W and 4.2W for Class 1 and 7.0W and 7.4W for Class 2. The columns are far enough apart there that 7 and 6.49 do not get confused. The 15.4 figure does get read as a device figure, because one vendor page presents it as one.
Above PoE+ the pattern holds and the gaps widen. The Catalyst 9400 chapter states that the “IEEE 802.3bt standard enables delivery of up to 90W to a powered device, over four pairs of Category 5e and above cables” — here Cisco does attach the figure to the device. Ubiquiti's article, staying with its own convention, gives the same tier as 60W for type 3 and 100W for type 4 at the PSE. NETGEAR's table splits it again into a 25.5W-40.0W device range against a 45.0W minimum switch-port output, and a 40.0W-51.0W device range against a 60.0W minimum switch-port output. Three documents, three framings, and the only way to compare them is to ask each one which end of the cable it is standing at.
A Diagnosis Order That Uses the Right Number
When a powered device will not come up, work the numbers in this order rather than starting with the cable.
Read the switch's per-port allocation, not the marketing figure. On a Cisco chassis that is the allocated power the port reports, in milliwatts. If it reads 15400 and the device is Class 3, the switch is behaving correctly and the device's 12.95-watt entitlement is intact.
Compare the device's stated need against the device column rather than the port column. If the device asks for more than 12.95 watts, it is not an 802.3af device regardless of what class it advertises.
Sum the reservations, not the draws, against the switch budget. Count 15.4 watts per Class 3 port unless you have configured a lower static value. This is the step that explains a port that worked until the day someone added one more camera.
Check the cabling limits the documents actually state: Cat.5 or above, and not over 100 metres. Those are the conditions under which the class ranges were calculated in the first place.
Confirm the port is eligible. The Catalyst 6500 chapter carries a restriction that costs people an afternoon: “PoE is supported only on Layer 2 switchports.” A port that has been converted to a routed interface supplies no power, and the symptom is indistinguishable from a dead injector. Configuration state that silently disables a feature is the same failure shape as a port-forwarding rule that saved correctly while the WAN side made it inert.
When you escalate, quote the class row rather than the headline figure. A support ticket that says the port allocates 15400 milliwatts while the device is rated at 12.9 watts gives a vendor engineer both columns and skips the round trip where they ask which one you meant.
When This Doesn't Apply
Passive PoE is outside all of this. Ubiquiti's article lists 24V, 48V and 54V passive options, and its entries say only that each provides power at that voltage: no class, no device-side figure and no port-side allocation appear anywhere on the page. Detection and classification signatures were not collected for this piece, so the absence recorded here is an absence in the documents. The 15.4 and 12.95 figures have nothing to say about a passive injector, and connecting a standards-based device to one is a different question entirely.
Pre-standard implementations are also outside it. The Catalyst 6500 chapter lists Cisco prestandard inline power at 10 W at the PSE, a figure that predates the class table and does not map onto it.
Platform support limits the upper tiers regardless of what the standard permits. The Catalyst 9400 chapter states that only two listed line cards support IEEE 802.3bt, and then qualifies the ceiling it has just set: “802.3bt Type 4 PSE is not supported. However, you can power-up a Type 4 PD through a power demotion to 60W.” Unsupported here means demoted, not dark. A standard's ceiling is not a product's ceiling, and the product documentation wins.
Finally, this reasoning is about switch ports. A midspan injector between a non-PoE switch and the device introduces its own supply figure and its own budget, and the switch's stated PoE budget does not describe it.
What Would Invalidate This
The IEEE 802.3 clause text was not read. IEEE 802.3af, 802.3at and 802.3bt clause language is not openly retrievable, so every standard-side figure here comes from manufacturer documentation restating it. If the clause text assigns 15.4 watts to the device side, the five documents above are wrong together and so is this piece.
None of the five documents prints a loop-resistance figure, a per-metre loss figure, or a derivation of 12.95 from cable resistance. The claim made here is only the one NETGEAR prints: the ranges were calculated at the 100-metre maximum. The physical derivation is inferred and would need the clause text to confirm.
The five documents collected go as far as the class figures and the allocation mechanism, and the behaviour of a port while a Class 3 device draws between 12.95 and 15.4 watts sits outside their scope. That band is where an intermittent power cycle would live, and this piece asserts no behaviour there.
Detection and classification signature resistances, and the port voltage ranges for active PoE, were not collected. Whether a particular camera or access point model is Class 3 or Class 4 was also not checked against that model's own datasheet, and a device that misclassifies itself would produce the same symptom from a different cause.
Comments
Post a Comment