Power Supply Efficiency Ratings and Thermal Derating: Why a 92% Supply Gets Used at 70%

|Power Supply Knowledges|10/22/2026|11.7 min|
Table of contents
Share Post

A 150 W supply rated at 92% efficiency died after six months inside an outdoor light box. The customer’s first response was that the efficiency figure was inflated. The second was that the build quality was poor.

We measured it. Efficiency came in at 91.7%. Not inflated.

The actual problem: it had been running at full load in a 50 °C ambient, and its derating curve starts rolling off at 45 °C. At 50 °C, continuous output capability was down to 78%.

The 150 W on the nameplate was real — measured at 25 °C ambient with good ventilation. The site has no 25 °C.

This article covers two things: what each of the efficiency programmes (80 PLUS, DoE Level VI, ErP Lot 6, CoC Tier 2) actually regulates and how little they overlap; and how to read a derating curve, measure case temperature properly, and understand why a one-point efficiency gain isn’t about the electricity bill.

80 PLUS and DoE Level VI don’t cover the same products

Procurement specs routinely demand “80 PLUS Gold” on an LED driver. The two have nothing to do with each other.

80 PLUS certifies internal ATX power supplies — the unit that plugs into a motherboard in a desktop or server. It’s voluntary, promoted by a US environmental organisation, tested at 10% / 20% / 50% / 100% load, with tiers from White (80%) to Titanium (94% at 50% load).

DoE Level VI regulates external power supplies (EPS) — the adapter at the far end of the cable. It’s defined in 10 CFR Part 430 Subpart B Appendix Z, administered by the US Department of Energy, and mandatory since 10 February 2016. Efficiency is the average of 25% / 50% / 75% / 100% load points, with a separate no-load power cap. Compliant products must be registered in the DOE Compliance Certification Database and marked with a Roman numeral “VI”.

The Level VI limits for single-voltage AC-DC basic-voltage models:

Nameplate output (Pout) Minimum average active efficiency Max no-load power
Pout ≤ 1 W ≥ 0.5 × Pout + 0.16 ≤ 0.100 W
1 W < Pout ≤ 49 W ≥ 0.071 × ln(Pout) − 0.0014 × Pout + 0.67 ≤ 0.100 W
49 W < Pout ≤ 250 W ≥ 0.880 ≤ 0.210 W
Pout > 250 W ≥ 0.875 ≤ 0.500 W

Note the logarithmic segment between 1 W and 49 W: a 24 W adapter faces a higher percentage bar than a 6 W one. And note the no-load caps — 100 mW for most adapters. That’s where designs actually fail. Teams optimise full-load efficiency and then blow the limit with a leaky standby controller, a bleed resistor or a power LED.

ErP Lot 6 Tier 2 is the EU equivalent: Regulation (EU) 2019/1782, in force since 1 April 2020. Numerically close to Level VI, documented through the CE technical file and declaration of conformity, with EPREL registration for some products.

CoC Tier 2 is a Code of Conduct published by the European Commission’s Joint Research Centre. Voluntary — and stricter than both mandatory regimes, typically one to several points higher on efficiency with lower no-load limits. Why design to a rule nobody enforces? Because green public procurement, eco-labels and large-brand specifications frequently list it as a purchase condition.

GEMS MEPS is mandatory in Australia and New Zealand and requires a registration number before shipment.

So which one applies to an LED driver?

None of them, cleanly. The DoE EPS definition explicitly excludes low-voltage luminaire drivers and battery chargers. LED drivers run on a different track:

  • Safety: IEC/EN 61347-2-13 for LED drivers, plus IEC/EN 62368-1 depending on construction
  • EMC: EN 55015 (conducted and radiated emissions), EN 61000-3-2 (harmonic current), EN 61547 (immunity)
  • Energy: in the EU, Regulation (EU) 2019/2020 covering light sources and separate control gear — not ErP Lot 6
  • Performance: IEC 62384

The rule: a plug-in adapter with a cord that goes into a wall socket runs DoE Level VI and ErP Lot 6. A driver built into a luminaire or a control cabinet runs 61347-2-13 and (EU) 2019/2020. Neither substitute for the other. We break the rating systems down side by side in power supply efficiency ratings explained: 80 PLUS, DoE Level VI, ErP and what buyers need — worth reading if you’re writing a spec that mentions Level VI on an LED driver.

And one more signal worth noting: if a supplier quotes “80 PLUS Gold” on an LED driver, that datasheet was almost certainly copied from a PC power supply template.

A one-point efficiency gain saves 2 watts. It decides whether capacitors last 5 years or 9.

Do the arithmetic on 150 W:

  • 88% efficient → input = 150 / 0.88 = 170.5 W, losses 20.5 W
  • 89% efficient → input = 150 / 0.89 = 168.5 W, losses 18.5 W

One point saves 2 W. At 0.12/kWh and 4,000 operating hours a year, that's **0.96 annually.** Over a project lifetime, that number is noise. The “efficiency saves money” pitch simply doesn’t hold up at these power levels.

But where did those 2 W go? Into the enclosure, as heat.

A convection-cooled aluminium supply runs roughly 2–3 K/W thermal resistance. A potted IP67 plastic one can reach 4–6 K/W. Those 2 W translate to:

  • IP20 aluminium: 4–6 °C cooler
  • Potted IP67: 8–12 °C cooler

Apply the 10 °C rule for electrolytic capacitors — every 10 °C halves life — and in potted construction, 2 W is worth 1.7–2.3× the capacitor life. A supply warrantied for five years becomes either eight or three.

Efficiency’s real value isn’t the electricity. It’s the temperature.

In a sealed, high-IP, fanless structure, every efficiency point is buying life directly — the mechanism is laid out in IP20 vs IP65 vs IP67: choosing an enclosure rating, where a potted IP67 housing runs 21 K hotter than an IP20 aluminium one on identical losses. That’s also why a high-efficiency design is worth paying for on an outdoor IP67 unit and not worth paying for on a well-ventilated indoor IP20 one. The sealed-housing version of this problem is covered in thermal management strategies for power supplies, and it’s the reason our 300 W slim IP67 waterproof supply is specified with a wider thermal margin than the same rating in an IP20 housing.

And full load isn’t the efficiency peak

The efficiency curve is an inverted U, peaking somewhere around 50–75% load. At full load, conduction losses (I²R) rise and efficiency sits 1–3% below the peak. At light load, fixed losses — controller quiescent current, snubbers, core loss — dominate, and 10% load can be 8–15% below peak.

So sizing at 100% is a double loss: you run at a worse efficiency point and at the highest case temperature. Leaving 20% headroom — running at 80% load — usually lands near the efficiency peak while holding temperature down.

Derating curves: the horizontal axis is ambient temperature, not case temperature

This is the single most common derating mistake.

The horizontal axis on a derating curve is ambient temperature — the temperature of the air around the supply. Plenty of engineers point an IR gun at the case, read 70 °C, and look up “70 °C” on the curve. That’s the wrong axis, and it produces an absurdly conservative answer.

The reverse error is just as common: using the Tc measurement point temperature as if it were ambient, which under-derates.

Measure two values and check them against two different limits:

Parameter How to measure Checked against
Ambient temperature Thermocouple suspended about 50 mm from the supply’s air intake, clear of other heat sources Derating curve (sets available output power)
Case temperature Tc Thermocouple bonded at the Tc point marked in the datasheet (usually near the transformer or on the heatsink) Max case temperature limit (sets life and safety)

A supply showing 70 °C case temperature at full load in a 25 °C room is entirely normal, provided the datasheet’s maximum case limit is 85 °C or 90 °C. Those two numbers are different quantities and shouldn’t be converted into each other.

What a typical curve looks like

The usual shape: no derating up to 45–50 °C ambient, then a linear fall, reaching 50–60% of rated output at 70 °C.

Applied to the opening example: a 150 W supply at 50 °C ambient delivers about 117 W continuous (78%); at 70 °C it manages 75–90 W.

That’s the whole explanation for “rated 150 W but only usable at 100 W.” The nameplate isn’t lying — it’s just measured at 25 °C.

Derating isn’t conservatism. It’s the capacitor core temperature boundary.

Engineering teams often read derating curves as the manufacturer covering itself. The curve’s shape is actually drawn by one physical constraint: the core temperature ceiling of the output electrolytic capacitor.

Electrolytic life is a function of core temperature, which is ambient plus self-heating from ripple current plus conduction from neighbouring hot parts. A 105 °C / 5,000 h capacitor needs its core held below roughly 75 °C to reach 30,000 hours.

Every point on a derating curve is the solution to: at this ambient, how far must output be reduced to keep core temperature at or under the design ceiling? It isn’tpadding. It’s a boundary.

Once you see it that way, the correct procedure follows:

  1. Establish your real maximum ambient — not the monthly average, but the peak inside a sealed enclosure on a summer afternoon
  1. Read the corresponding percentage off the derating curve
  1. Divide your load by that coefficient to get the required rated power — you don’t size to the load, you size to load ÷ derating factor

Example: 120 W load, peak ambient inside a sealed box 55 °C, derating factor ≈0.72 → required rating ≥ 120 / 0.72 = 167 W. Specify a 200 W unit.

Pick a 150 W unit for a 120 W load and you’ll get thermal shutdown or early failure when summer arrives.

Three environmental variables that get missed

Altitude. Natural convection depends on air density, which falls with altitude. Most manufacturers apply:above 2,000 m, derate a further 2–5% per 1,000 m. Plateau projects need that applied on top of the temperature coefficient. Altitude also affects clearance distances through the IEC 60664-1 correction factor — an independent derating path that’s easy to forget.

Side-by-side mounting. A supply installed alone and one in a row of six are in completely different thermal situations. Two neighbours raise the ambient seen by the unit between them by 10–15 °C — it isn’t being cooked by its own losses, it’s being cooked by the ones next to it. Practical rule: leave ≥50 mm above and below convection-cooled units and ≥15–20 mm between units mounted side by side. If you must pack them densely, re-read the derating curve against the resulting local ambient.

Sealed enclosures. Inside a control cabinet, light box or in-ground enclosure, temperatures run well above the outside air. A metal light box in direct summer sun can sit 20–30 °C above ambient. That number has to be measured, not guessed — don’t look up “35 °C outdoor” on the curve. Mounting method changes the thermal path just as much as spacing does: DIN rail power supply installation and wiring covers how rail mounting inside a cabinet affects both heat dissipation and the derating you should apply.

Measuring case temperature on an installed unit: three steps

1. Find the hot spot. Sweep the housing with an infrared camera. Without one, probe point by point: above the transformer, on the housing over the MOSFET heatsink, near the output capacitors, and on the top surface.

2. Bond a thermocouple at Tc. Use a fine-gauge K-type held with thermal adhesive or high-temperature tape, placed at the Tc point the datasheet marks. Unmarked? Put it on the housing surface beside the transformer.

3. Run at full load until thermal equilibrium. Switch-mode supplies have a thermal time constant of 1–2 hours; take the reading at 4 hours. Measuring at 30 minutes reads noticeably low — a lot of “seems fine” sites are simply measured too early.

Then check two things: is Tc under the datasheet’s maximum case limit (commonly 85 °C or 90 °C), and does the load sit within what the derating curve allows at your measured ambient?

FAQ

Which is stricter, 80 PLUS or DoE Level VI? Not comparable. 80 PLUS is voluntary and applies to internal PC ATX supplies, tested at 10/20/50/100% load. DoE Level VI is a federal mandate covering external adapters, averaging 25/50/75/100% load with a no-load cap. An adapter can’t earn 80 PLUS; a server PSU isn’t bound by Level VI.

Does an LED driver need DoE Level VI? Usually not. The DoE EPS definition excludes low-voltage luminaire drivers. LED drivers follow IEC/EN 61347-2-13 for safety, EN 55015 for EMC, and (EU) 2019/2020 for separate control gear in Europe. If a supplier hands you a Level VI certificate for an LED driver, it probably belongs to a different product line.

Is one efficiency point worth paying extra for? Depends on the structure. In a well-ventilated indoor IP20 installation, no — the power saving is trivial. In a sealed potted IP67 housing, yes — one point is 8–12 °C off the case temperature, which by the 10 °C rule is 1.7–2.3× the capacitor life.

Why won’t the supply carry its rated load in summer? Because nameplate power is rated at 25 °C ambient, and derating typically begins at 45–50 °C. Use the peak ambient inside your enclosure in summer to read the curve — that’s the real available power.

What temperature goes on the horizontal axis of the derating curve? Ambient — the air around the supply. Not case temperature. Measure case temperature separately and check it against the maximum case limit. They’re independent; don’t convert between them, and don’t read an IR gun’s case reading onto the derating curve.

CoC Tier 2 is voluntary — why bother? Because large buyers require it. Green public procurement, eco-labels and major brand specifications routinely list CoC Tier 2 as a purchase condition. And today’s voluntary tier tends to become tomorrow’s mandatory floor, so designing to CoC leaves headroom.

Can I skip derating and just add a fan? You can, but run the reliability numbers first. A fan is a mechanical wear item rated 30,000–70,000 hours — shorter than the electrolytic — and in any dusty environment it’s always the first part to fail. Adding a fan trades one reliability problem for another. Solve it with convection if you can.

Three extra numbers to ask for

Next time you get a datasheet, confirm three things alongside power and efficiency: the knee temperature of the derating curve (40, 45 or 50 °C), the percentage still available at 70 °C, and the maximum permitted case temperature Tc.

Those three numbers decide how many watts the supply actually delivers on your site. The number on the nameplate is only true in a 25 °C laboratory.

Not sure what derating factor your site needs? Send the load power, mounting arrangement (sealed or vented, single unit or in a row), altitude and summer peak ambient temperature to engineering — we’ll work back through the derating curve to a recommended rating and supply the test report format to verify it on site. Certification documents and TDS sheets are in the download centre, and the safety certifications required for commercial LED drivers list is worth checking against your target market before you write the spec.

SANPU

Stay in the loop

Subscribe to our free newsletter.