Walk into a Tuesday morning retrofit and you see the same thing: Cat 5e pulled in 2011, zip-tied tight every eighteen inches, and a spec sheet asking you to hang a PTZ camera and three access points off the far end. Nobody replaced the cable. Somebody replaced the switch.
That is where PoE cable selection stops being academic. Power over Ethernet runs happily on almost anything until you demand real wattage over real distance inside a real bundle. Then gauge, copper quality, bundle size and connector grade decide whether the link survives the next five summers.
The Four Types, and Where Type 4 Breaks the Old Assumptions
Most installers know the headline wattages. Fewer remember that the wiring scheme changed.
| Type 1 (802.3af) | Type 2 (802.3at) | Type 3 (802.3bt) | Type 4 (802.3bt) |
|---|
Standard year | 2003 | 2009 | 2018 | 2018 |
|---|
Max PSE output | 15.4 W | 30 W | 60 W | 90 W |
|---|
Guaranteed power at device | 12.95 W | 25.5 W | 51 W | 71.3 W |
|---|
Pairs carrying power | 2 | 2 | 4 | 4 |
|---|
Max current per pair | 350 mA | 600 mA | 600 mA | 960 mA |
|---|
Typical endpoints | VoIP phones, fixed cameras | PTZ cameras, access points | digital signage, LED lighting | laptops, thin clients |
|---|
Type 1 puts 15.4 W on the wire and guarantees 12.95 W at the device. The gap is cable loss — 16% of your power gone before it does any work. Type 2 raises the guarantee to 25.5 W. Both use two pairs for power, the same pairs that carry data on 10/100 links or alternate on Gigabit.
Type 3 and Type 4 rewrote the wiring. They push power down all four pairs, which is why you will see them called 4PPoE. Type 3 delivers 51 W at the device at up to 600 mA per pair. Type 4 reaches 71.3 W at 960 mA per pair.
Doubling the conductors means each one carries less current for the same wattage. But you are now pushing DC through the pairs that carry your Gigabit or 10 Gigabit data. DC resistance unbalance stops being a lab curiosity and becomes a field problem — retransmits, and links that drop for no reason anyone can find.
Heat Is the Spec Nobody Reads
Power entering a cable does not all leave the other end. Some becomes heat inside the jacket, and heat does two things: it raises insertion loss, and it shortens insulation life.
One 90 W link sheds a handful of watts over 100 m. Bundle forty in a riser with no airflow and the middle cable sits in its own heat plus everyone else’s. The switch does not care. The cable does.
TIA wrote a bulletin about it. TSB-184-A, approved in March 2017 by TIA Subcommittee TR-42.7, is titled *Guidelines for Supporting Power Delivery Over Balanced Twisted-Pair Cabling*. It adds to the ANSI/TIA-568 infrastructure requirements rather than replacing them. Its recommendations:
- Leave cables unbundled where you can. When you cannot, keep bundles small.
- Limit bundles to 24 cables to control temperature rise from installation conditions, high ambient temperature, 26 AWG conductors and higher currents.
- Use Category 6A or higher for new PoE installations.
- Watch pair-to-pair DC resistance unbalance.
That 24-cable figure gets misquoted constantly. It is not a hard ceiling for every cable at every current — it is a conservative default aimed at the awkward cases.
One more lever: UL’s Limited Power (LP) listing evaluates jacket material against PoE temperature rise, corrected for a 45 °C ambient. LP on the jacket means the bundled-heat problem was tested, not just data performance.
The number that governs everything: 15 °C
TIA recommends capping temperature rise above ambient at 15 °C, and commercial cable is typically rated to 60 °C. Work backwards from those two numbers and you get real bundle limits. Tested at 1,000 mA per pair:
Cable | Allowable bundle size at 1,000 mA/pair (15 °C rise) |
|---|
Cat 5e | 52 cables |
|---|
Cat 6 | 64 cables |
|---|
Cat 6A | 74 cables |
|---|
Push to a 100-cable bundle and the ceiling drops hard. Leviton tested 100-cable bundles across categories and found Cat 6A holding 865 mA per pair at the 15 °C limit, while the lower categories gave up sooner. The category you buy decides how many cables you can crowd into a J-hook: twenty-four cables of Cat 6A is comfortable, sixty cables of Cat 5e is not.
What going over the limit costs you
Leviton priced the penalty in copper. They coiled 90 m of cable, terminated both ends, and applied 600 mA per pair across all four pairs at 60 °C ambient. Unshielded cable failed insertion loss and needed more than 9 m cut off before it passed. Foil-shielded cable passed at the full 90 m, and so did a cable with a metallic isolation wrap. Nine meters off a 90 m run is 10% of your reach — on a link budgeted at 100 m, that is the difference between a passing certification and tearing out a completed run.
Gauge: 23 AWG, 24 AWG, and the Thin Cords People Keep Buying
Cat 5e is commonly 24 AWG. Cat 6 and Cat 6A are usually 23 AWG. Patch cords show up at 26 AWG and even 28 AWG — thinner, more flexible, measurably worse under load.
Resistance is the reason, and the physics is simple. IEEE’s own 802.3af working material lists DC resistance for a 24 AWG conductor at 25.7 Ω per 1,000 ft, roughly 8.4 Ω per 100 m. A 23 AWG conductor is about 26% larger in cross-section, and resistance scales inversely with area — so expect roughly a fifth less resistance over the same length.
At 350 mA that gap is irrelevant. At 960 mA it shows up twice: as heat inside the jacket, and as voltage lost before power reaches the device. Voltage that never arrives is wattage you buy over again at the switch.
This is why TSB-184-A singles out 26 AWG. Thin cords are fine for a two-foot run between switch and patch panel. Put one on a 90 W camera run and you added resistance exactly where current is highest. One rule covers it: 23 AWG solid copper for anything carrying Type 3 or Type 4 power. The cost difference over 305 m is small. The call-back cost is not.
Pure Copper vs CCA: Where Cheap Cable Bites You
Copper-clad aluminum is the most expensive cheap cable on the market. The copper is a plating; the conductor underneath is aluminum.
Aluminum has roughly 55% higher resistance than copper at the same cross-section. Under PoE that becomes heat, voltage drop, and power burned in the cable instead of delivered to the device. CCA also fails the standards outright — it does not meet ANSI/TIA performance requirements.
Aluminum is brittle, too, and a tightly bent CCA conductor can crack its core inside the jacket where nobody will see it. The link tests fine on day one and goes intermittent in month eight.
Ask for a DC resistance test report per reel. Any real manufacturer can produce one. If the answer is a marketing PDF, you have your answer.
Shielded or Unshielded Under Load
The honest answer: unshielded is usually correct, and shielding is often sold for the wrong reason.
Shielding protects against alien crosstalk and electromagnetic interference. It does not lower a bundle’s temperature. What the Leviton test showed is narrower: foil-shielded cable held insertion loss at 90 m under 600 mA and 60 °C where unshielded cable did not. The benefit is dielectric behavior at high temperature, and it exists only if the shield is terminated. An ungrounded foil wrap is an antenna, not a barrier, and can perform worse than plain unshielded cable.
Decide shielding from the environment — high-EMI industrial floors, imaging suites, pathways shared with VFD motor feeds — then bond it properly per the manufacturer. If you are not going to ground it correctly, do not buy it.
Connectors: Arcing, Pitting, and the Tests to Ask About
Every time someone unplugs a patch cord while the link is live, a small arc jumps between the contacts. Not dangerous. Cumulative. UL describes the mechanism plainly: unmating under electrical load causes arcing, which corrodes contact material until the connection degrades, and the connector eventually needs repair or replacement.
Two IEC test schedules apply, and it is worth being precise about which:
- IEC 60512-99-001 — connectors in twisted-pair cabling with remote power, the PoE and PoE+ era.
- IEC 60512-99-002 — the same method at increased current, representing Type 3 and Type 4 conditions.
Belden notes that in a successful IEC 60512-99-001 test, contact resistance change must stay under 20 mΩ after 100 mating cycles at 600 mA. Leviton’s tear-down found pitting after just 25 cycles with current flowing, and more after another 25 with reversed polarity.
Specify jacks and plugs tested to IEC 60512-99-002 for any 802.3bt deployment, and check the gold plating — TIA calls for 50 µm on the contact tines. Thin plating plus high current is a short conversation.
Field Testing: Check DC Resistance Unbalance
This is what separates a certified job from a job that mostly works.
Power is delivered as a common-mode voltage on each pair, so current splits evenly between the two conductors only if their DC resistance matches. Any imbalance distorts the data signal and produces bit errors, retransmissions, and links that fail under load.
Type 3 and Type 4 added a second problem. Because power now flows on all four pairs, excessive unbalance *between pairs* causes trouble, not just within a pair. A datasheet cannot save you — inconsistent terminations cause most of the imbalance found in the field, and only a test finds it. Certify unbalance within pair and between pairs, and treat marginal terminations as failures. Reterminating a jack costs minutes. Chasing an intermittent camera link costs a truck roll per visit.
A Buying Checklist
- Category: Cat 6A or better for new Type 3 and Type 4 installs. Cat 6 is defensible for retrofits where bundles stay small.
- Gauge: 23 AWG solid copper for powered runs. Reserve 26 AWG and 28 AWG for patch cords.
- Conductor: 100% solid bare copper, with a per-reel DC resistance report.
- Bundle size: Under 24 cables, or verify against the 15 °C table for your category.
- Ambient and pathway: Add ambient and rise. If the total approaches 60 °C, resize the bundle or move it, and match the jacket rating to the space — plenum for air-handling, riser for shafts.
- Connectors: Tested to IEC 60512-99-002 for 802.3bt, with 50 µm gold on the tines.
- Shielding: Chosen from the environment and bonded correctly. Not bought as a heat fix.
- Field test: Certify DC resistance unbalance within pair and between pairs, not just wiremap and length.
The Short Version
PoE did not change what a good cable is. It changed how fast a bad one shows itself.
Buy 23 AWG solid copper in Cat 6A, keep bundles small, keep them out of hot pathways, specify connectors that survived the arcing tests, and certify resistance unbalance instead of trusting wiremap. Do those five things and Type 4 power becomes boring. Skip any of them and you will be back on that ladder.
Sources
