PoE voltage drop and bundle heat field guide

PoE limits: delivered power and bundle heat, not just cable length

Power over Ethernet sends DC power and data down the same twisted pairs, and two things limit how far and how hard you can push it: the power that actually reaches the device after the cable takes its share, and the heat a loaded bundle builds up. The 100 m channel length everyone quotes is a data limit. It is not the power limit.

The cable has resistance, and that resistance does two things at once. It drops voltage, so the powered device sees less than the switch put on the line. And it turns the lost voltage into heat, which is fine for one cable in open air and a real problem for the center cable of a tight bundle in a hot ceiling. A run can pass a cable certifier for length and still starve a Type 4 camera, or cook a bundle past its temperature rating, with the link showing green the whole time.

So a PoE check is two checks. Will the device get enough power at the far end, and will the cable stay inside its temperature rating where it is bundled. Length is the easy one. These are the two that get missed.

PoE power classes: 802.3af, 802.3at, and 802.3bt

PoE comes in four types across three IEEE amendments, and the number that matters is the lower one: the power guaranteed at the device, not the power the switch is rated to source. The gap between them is the loss the standard sets aside for the cable.

Type 1 (IEEE 802.3af) sources 15.4 W and guarantees 12.95 W at the device over two pairs. Type 2 (802.3at, PoE+) sources 30 W for 25.5 W delivered, still two pairs. The 802.3bt amendment added four-pair power: Type 3 sources 60 W for 51 W delivered, and Type 4 sources up to 90 W for 71.3 W delivered, with each pair carrying close to 960 mA at full Type 4 load. Confirm the exact class and delivered figure against the device's listing, because manufacturer marketing names (PoE, PoE+, PoE++, UPOE, Hi-PoE) do not map cleanly onto the IEEE types.

Why does a PoE device see less power than the switch sends?

The device sees less because the twisted pairs have resistance, and current flowing through that resistance drops voltage and burns power as heat before it ever reaches the load. PoE runs at a nominal 44 to 57 VDC, and the standard budgets the channel resistance, around 12.5 ohms for a 100 m channel, as loss the design has to live with.

Two losses come out of the same resistance. The voltage drop is current times the loop resistance, so a Type 4 device pulling close to 960 mA per pair over a long, warm run can lose several volts on the way out. The power burned in the cable is current squared times resistance, which is why heat climbs fast as you load the pairs harder. Double the current and the cable heat goes up four times. That is the same I-squared-R loss that limits any conductor, just at low voltage where there is less headroom to give.

This is why the device's guaranteed power is lower than the switch's rated power. The standard already subtracted the worst-case channel loss. Push past worst case with a longer run, thinner conductors, or a hot bundle and the device can fall below its class even though the switch is doing its job.

PSE

Power sourcing equipment, the switch or midspan that puts power on the line

PD

Powered device, the camera, AP, or controller drawing power at the far end

Loop resistance

The round-trip resistance of the conductors carrying current, the source of both the drop and the heat

The 100 m channel and where the power budget goes

The 100 m Ethernet channel is built as 90 m of solid horizontal cable plus 10 m of stranded patch cord, with the connectors at each end counted in. The TIA channel resistance figure of about 12.5 ohms comes from that model: 90 m of 24 AWG solid run warm, plus the patch cords, plus four connector interfaces. Power has to get to the device across all of it.

Stranded patch cord is the quiet cost in that budget. Stranded conductors have higher resistance than solid for the same gauge, and the patch cords are often a thinner gauge than the horizontal, so 10 m of patch can take a disproportionate slice of the loss. Two long patch cords at each end, in 26 or 28 AWG, will move a marginal run from passing to failing.

The horizontal run is where you have the most control at design time. The patch cords are where the loss sneaks in later, because someone swaps in whatever is in the bin. On a high-power run, the patch cords are part of the power calculation, not an afterthought.

How far can high-power PoE run before it falls short?

Data goes 100 m. Full power does not always come with it. The link will train and pass at 100 m, but a Type 4 device pulling its full 71.3 W at the end of a warm 90 m run in 24 AWG can sit right at the edge of its delivered-power budget, and anything that adds resistance pushes it over.

The variables that decide it are the ones that add loop resistance: length, conductor gauge, temperature, and the patch cords. A Type 1 or Type 2 device at 12.95 or 25.5 W has plenty of margin at 100 m. A Type 4 device at 71.3 W does not, and that is the run where distance and power fight each other. The practical line is that high-power 802.3bt near full channel length is the case to actually calculate, not assume.

When the number is tight, you have four moves: shorten the run, go to a heavier gauge cable, drop in a midspan or PoE extender to re-power partway, or power the device locally and use the cable for data only. Pick the one the building allows. The wrong move is to install it long and thin and hope the device boots.

Conductor gauge: 24, 23, and 22 AWG

Gauge is the most direct lever on PoE power and heat, because a thicker conductor has lower resistance, which cuts both the voltage drop and the heat the cable makes under load. Cat5e is typically 24 AWG. Cat6 and Cat6A are commonly 23 AWG, one step heavier, which drops DC resistance by roughly 20 percent. High-power cables run 22 AWG to push that further.

For a long Type 4 run, gauge is usually the cheapest fix on the table. Going from 24 to 23 AWG buys back resistance across the whole length, which shows up as more volts at the device and less heat in the bundle. The catch is the patch cords, which are stranded and often two gauges thinner than the horizontal, so a heavy horizontal run still loses ground at the ends if the cords are 26 or 28 AWG.

Using 24 AWG for a long, high-power run is the mistake that looks fine until the device is on. It passes the length test, then it underpowers or the bundle runs hot. On Type 3 and Type 4 runs near full distance, 23 AWG or heavier is the conductor that holds up.

Cat5e, Cat6, or Cat6A for PoE: which runs cooler?

Cat6A runs coolest of the three for the same PoE load, because its heavier conductors and larger cross section carry the current with less resistance and shed heat better in a bundle. Cat6 is in the middle. Cat5e, on thinner 24 AWG, makes the most heat and the most voltage drop under high power, which is why it is a poor choice for Type 3 or Type 4.

The category does not change the IEEE power class. A device gets its 51 or 71.3 W from the standard regardless of cable. What the category changes is whether the cable delivers that power without dropping too much on the way and without overheating where it is bundled. On a long, high-power run that is the whole question.

For new high-power work, Cat6A is the default that keeps both voltage drop and bundle heat in hand, and shielded constructions help a little with heat by giving it another path out of the cable. Reusing existing Cat5e for PoE+ at short length is usually fine. Pushing existing Cat5e into Type 4 at distance is where the trouble starts.

Bundle heat and the 15 degree C rise

Every powered pair makes heat, and bundling cables traps it. The cable in the center of a tight bundle cannot shed heat to the air, so it runs hotter than the ones on the outside, and the whole bundle's temperature climbs with the number of powered cables and the power on each. TIA TSB-184-A frames the limit as a temperature rise: keep the center cable's rise to about 15 degree C over ambient at full PoE.

That rise stacks on the ambient, and the sum is what has to stay under the cable's temperature rating. A bundle in a 20 degree C ceiling has room. The same bundle in a 45 degree C rooftop or attic plenum starts close to the rating before any PoE heat is added, so the allowable bundle gets much smaller. This is why bundle guidance is always tied to ambient, not a flat cable count.

Heat also feeds back into power. A hotter conductor has higher resistance, so a hot bundle both pushes the cable toward its rating and drops more voltage to the device at the same time. The two failures, overheating and underpowering, show up together on the worst runs.

How many PoE cables can I bundle?

There is no single number, because the answer rides on ambient temperature, conductor gauge, the power per cable, and whether the bundle is in open air, tray, or conduit. TIA TSB-184-A gives the bundle counts that hold the center cable to a roughly 15 degree C rise, and as a working rule it points at limiting bundles to about 24 cables in a typical conduit or tray pathway at full power. Hotter ambient or thinner conductors push that count down.

The NEC handles the same physics as an ampacity question. Table 725.144 sets the allowable current per conductor against bundle size and conductor gauge (26, 24, 23, and 22 AWG) at a 30 degree C ambient, with correction factors applied for hotter spaces. There is a practical exception: the table does not apply where the conductors are 24 AWG or larger and the source delivers no more than 0.3 A per conductor, which covers a lot of low-power PoE but not Type 4.

When a bundle is too large, you do not have to redesign the run. Break it into smaller bundles, space them apart, or pull the high-power cables out of the dense center. Air around the cable is the cheapest cooling there is. Confirm the bundle limits against the adopted code edition and the cable manufacturer's PoE guidance, since both the table and the TSB get revised.

Conduit fill and PoE heat together

A crowded conduit is worse for PoE than for data alone, because conduit fill was written around pulling tension and insulation damage, not heat. Data cables make almost no heat, so a conduit packed to the fill limit is fine for plain Ethernet. Load those same cables with PoE and the conduit becomes an oven with no airflow, and the fill that was acceptable for data can cook the bundle.

Conduit makes the bundle-heat problem sharper in two ways. It packs the cables tight so the center conductors cannot shed heat, and it seals them away from moving air, so the heat that does build has nowhere to go. The center of a full PoE conduit is the hottest spot in the run, and it is the spot you cannot see or measure after the fact.

Plan a PoE conduit with headroom, not at the fill maximum. Leave it emptier than a data-only design would, keep the high-power runs out of the most crowded pipes, and treat a conduit already full of energized PoE as a place not to add more. Mixing a new high-power run into a packed, warm conduit is asking for the heat failure you cannot inspect.

Field example: a PTZ camera at the fence line

Site cameras are the classic PoE power problem, because they sit at the far end of the longest runs and the pan-tilt-zoom and heater models pull real power. A PTZ with a heater can be a Type 3 or Type 4 load, 51 to 71.3 W at the device, and it is parked 80 to 90 m out at the property line where the run is longest and often the hottest.

Run the case. A Type 4 PTZ at 71.3 W on 24 AWG at 90 m, in a sun-exposed conduit along a fence, is loading thin conductors hard, at distance, in high ambient, all at once. That is the run that boots fine on the bench and then resets at 2 p.m. when the conduit heats up and the camera's heater kicks on. The cable did not change. The conductor got hotter, its resistance rose, and the delivered power fell below what the camera needs.

The fixes are the usual four, ranked by what the site allows: pull 23 or 22 AWG instead of 24, shorten the run by moving the switch or a field enclosure closer, drop a PoE extender or midspan into a weatherproof box partway out, or power the camera locally and run the cable for data only. For a long fence line, a field enclosure with a switch is often cheaper than fighting the distance.

Wireless APs, Wi-Fi 6E and Wi-Fi 7

Access points climbed the power classes as the radios multiplied. An older single-radio AP ran happily on Type 1. A current Wi-Fi 6E or Wi-Fi 7 AP with multiple radios, more spatial streams, and USB or IoT ports can need Type 3 or Type 4 power to run every feature at once, and it will quietly drop features if it does not get it.

That is the AP trap: it boots and associates clients on PoE+ when its full feature set wants PoE++. It does not fail. It throttles, disables a radio, or shuts down the auxiliary ports, and the symptom is weak coverage or missing capacity that gets chased as an RF problem when it is a power problem. Check the AP's required class for the features you are actually turning on, not just the class that lets it boot.

APs also live above ceilings and in hot plenum spaces, so the bundle-heat side applies to them too. A tight bundle of high-power APs feeding a hot ceiling is the same derating problem as the camera bundle, spread across the building. Confirm the powered class against the manufacturer's full-power figure, then check the run and the bundle for it.

Access control and door controllers

Door controllers and access panels are lower-power than cameras or APs, but they are the runs where a reset is least acceptable, because the failure can be a door that locks or unlocks wrong. Most controllers, readers, and PoE locks fall in the Type 1 to Type 2 range, so power budget is rarely tight. The problem on these is reliability under heat and the shared bundle, not raw watts.

The run that bites is the controller sharing a hot bundle with high-power cameras. Its own draw is small, but it sits in a bundle whose temperature is set by the Type 4 cables next to it, so it inherits their heat and the resistance rise that comes with it. A door controller that drops out intermittently in the afternoon, in a bundle full of PTZ cameras, is reading the bundle's heat, not its own.

Where a door has to fail to a known state, the power path is part of the life-safety design, so the class, the run, and the backup get verified against the controller and lock listings. Keep these runs out of the hottest bundles, and do not assume a low-power device is safe just because its own watts are small.

Does the RJ45 connector heat up under 4-pair PoE?

Yes, the connector is a real heat source under high-power PoE, because every mated contact has a small contact resistance, and four-pair Type 3 and Type 4 power pushes current through all eight contacts at once. Current squared times that contact resistance makes heat right at the pins, and a worn, dirty, or loosely mated contact makes far more of it than a clean one.

The risk that gets attention is unmating under load. Pulling a plug while it is carrying full Type 4 current can draw a small arc at the contact as it breaks, which over many cycles pits the contact, raises its resistance, and makes it run hotter still. The 802.3bt connector guidance covers this with disconnect and contact requirements, which is why high-power PoE wants connectors and jacks rated for it, not bottom-bin parts.

In the field the connector shows up as the worst spot in a run. A termination that reads hotter than the cable, or a run that delivers worse than the cable length alone predicts, is usually a bad contact, not bad cable. Terminate clean, seat the plug fully, and on high-power runs use components rated for four-pair PoE.

DC resistance unbalance, the 4-pair failure you can test for

Four-pair PoE splits current across the pairs and across the two conductors in each pair, and it counts on that current sharing being even. DC resistance unbalance is when it is not: one conductor or one pair carries more than its share because its resistance is lower, so the overloaded conductor runs hotter and drops more than the design assumed.

The standard sets limits on it. Within a pair, the two conductors are held to a small resistance difference, and between pairs the 802.3bt limit caps the difference so the four pairs share Type 3 and Type 4 current evenly. Unbalance usually comes from a bad termination, a partially broken strand, or mixing conductor gauges in one channel, not from good cable run straight.

This is a test you can actually run. A cable certifier that measures DC resistance unbalance reads it directly, and on high-power PoE it is cheap insurance against a run that passes length and wiremap but cannot carry four-pair power evenly. When a high-power run misbehaves and the length tests clean, resistance unbalance is the next thing to check.

Why does my PoE device reset, underpower, or refuse to boot?

A PoE device that resets, runs in a reduced mode, or will not boot is almost always a power-delivery problem, not a dead device, and the cause is usually that delivered power fell below the device's class. The device negotiates a class with the switch, draws power, and when the voltage at its input sags under load below what it needs, it browns out and reboots, often in a loop.

Work it in order. Confirm the switch port actually supplies the device's class, since a Type 2 port cannot run a Type 4 device no matter how short the cable. Check the run length and gauge against the power the device needs, because a long thin run is the common starve. Look at the bundle and the ambient, since a device that is fine in the morning and resets in the afternoon heat is reading temperature. Then check the terminations and the patch cords, because a bad contact or a thin 28 AWG cord at each end drops power right where you do not expect it.

The tell that separates the causes is timing. A device that never boots is usually class or wiring. A device that boots and then resets under load, or only when it is hot, is a delivered-power and heat problem, and the fix is the same set of moves: heavier gauge, shorter run, midspan, or local power.

Extenders, midspans, and local power

When a run cannot deliver the power at the distance, you have a short list of fixes, and they trade cost against how much of the cable plant you keep. Heavier gauge is the first reach: pulling 23 or 22 AWG instead of 24 buys back resistance over the whole length and is cheap if the cable is not in yet. Shortening the run, by moving the switch or adding a field enclosure closer to the load, attacks the distance directly.

A PoE extender or midspan re-powers the line partway out. An extender repeats both data and power and can push a device well past 100 m, while a midspan injects power onto a data run that comes from a non-PoE switch. Both need their own power where you place them, so they want a weatherproof, ventilated enclosure, which is itself part of the cost.

Local power is the move when distance beats every other option. Run normal data to the device and power it from a nearby source, and the channel-power problem disappears. For a row of cameras at a far fence or a cluster of high-power APs in one wing, a small field enclosure with a local switch is often cheaper and more reliable than fighting voltage drop and heat over a long, thin run.

What to document

A PoE run that underpowers six months later is hard to diagnose without the record of what was designed, because the symptom, a device resetting, points at the device first and the cable last. Write down enough that the next person can see whether the run was ever right for the power it carries.

Capture the device's required power class and delivered watts, the cable category and conductor gauge, the one-way run length, the bundle count and pathway (open, tray, or conduit), the ambient at the worst point, any derating or TSB bundle assumption used, and whether an extender, midspan, or local power was added. If the run is near a limit on power or heat, write down which limit and how close, so a future change does not quietly push it over.

Common mistakes

• Assuming a run that passes the 100 m length test will also deliver full power to a Type 4 device.
• Ignoring bundle heat and packing high-power PoE cables tight with no airflow.
• Adding a new high-power run into a conduit already full of energized PoE.
• Using 24 AWG for a long Type 3 or Type 4 run instead of 23 or 22 AWG.
• Sizing the run to the device's marketing name (PoE++) instead of its IEEE class and delivered watts.
• Forgetting the stranded patch cords, which are thinner and add loss right at the ends.
• Treating ambient as room temperature when the bundle lives in a hot ceiling, attic, or rooftop conduit.
• Powering a high-power device through bottom-bin RJ45 parts not rated for four-pair PoE.

Field checklist

Standards and references

The power side is IEEE. The 802.3 standard defines the Ethernet channel and its 100 m length, and the PoE amendments define the types and power: 802.3af for Type 1, 802.3at for Type 2, and 802.3bt for Type 3 and Type 4 four-pair power. The delivered-power figures (12.95, 25.5, 51, and 71.3 W) and the channel resistance budget come from those amendments, so confirm a device's exact class and delivered watts against its listing.

The heat and cabling side is TIA and the NEC. TIA-568 and the ANSI/TIA cabling standards define the categories and the channel, and TIA TSB-184-A gives the PoE-specific guidance on bundle size and the roughly 15 degree C temperature-rise limit. In the NEC, the power-and-data transmission rules and Table 725.144 set allowable current per conductor against bundle size and gauge, with PoE generally run as a Class 2 power-limited circuit and premises-powered broadband falling under Article 840. The exact article, section, and table numbers shift between code cycles and the TSB gets revised, so confirm them against the adopted edition, local amendments, and the cable manufacturer's PoE documentation before you cite them on a submittal.

Where a device listing or manufacturer instruction sets a tighter requirement than the general guidance, the listing governs. Cite the standard that controls the point, and let the project specification and the equipment listing override a rule of thumb.

Units, terms, and conversions

PoE numbers move between watts, volts, amps, and ohms, and the cable world adds gauge and length units that read differently across a spec, a datasheet, and a drawing. Power is in watts at both the source (PSE) and the device (PD), and the gap between them is the cable loss. Run length is in meters in the IEEE and TIA world, where 100 m is the channel and 90 m is the horizontal, while a US drawing may give the same run in feet (100 m is about 328 ft, 90 m about 295 ft).

Conductor size is in AWG, where a lower number is a heavier conductor with less resistance. Temperature ratings and rise are in degree C in the cable standards, where a 15 degree C rise is about 27 degree F. Current per conductor and per pair, in milliamps, is what the bundle-heat and ampacity rules actually limit.

PSE / PD

Power sourcing equipment (the switch or midspan) and the powered device at the far end

Type 1 to 4

IEEE PoE classes: af (15.4 W), at (30 W), and bt Type 3 (60 W) and Type 4 (90 W) at the source

4PPoE

Four-pair PoE, the 802.3bt method that uses all four pairs for Type 3 and Type 4 power

Channel vs link

The link is the permanent horizontal cabling; the channel adds the patch cords and end connections

AWG

American Wire Gauge; a lower number is a heavier conductor with lower resistance

Temperature rise

The increase over ambient from PoE heating; TSB-184-A targets about 15 degree C for the center cable

Related tools

Calculators and readiness checks for this work

Compare your options

FAQ

People also ask

• What voltage drop is acceptable?
• What size wire for a 48A charger?
• What is a buck-boost transformer?
• What is a low-voltage system?
• What is a stress cone?