This database holds 280 sources, and getting there meant working through more than 488 products. 208 were checked and left out: most because no usable number could be found, a few because the product turned out not to exist at all, and a few more because no current price could be established for it. What follows is what that reading turned up.
A headphone amplifier has one job that can be written down: deliver so many milliwatts into so many ohms. That is two numbers. A surprising share of the industry publishes one of them, or neither, or publishes a third number that looks like power and is not.
There are 1312 power figures across the database, one per source per load. Each one is labelled, and the labels are not decoration — they are the difference between a manufacturer's claim and this site's arithmetic.
| Status | Figures | What it means |
|---|---|---|
| Published at that exact load | 528 | the maker printed this power, into this impedance |
| Between two published loads | 159 | output voltage interpolated between the two figures either side |
| Beyond the published loads | 429 | output voltage held at its last published value, which understates rather than flatters |
| From a maximum voltage only | 96 | the maker published a voltage and no power, so this is a ceiling, not a rating |
| From one load plus output impedance | 42 | the voltage-divider model, which is the right one for a high-impedance output |
| Not yet traced to a document | 40 | carried over from the material this database was first assembled from |
40% of all 1312 figures are the manufacturer's own, at the manufacturer's own load — though that count flatters itself, because the odd loads in this database (16, 33, 40, 50, 62, 100, 120, 250, 330, 470 ohms) exist only because a maker published a figure there, so they are published by construction. At the four loads every source page actually shows, 346 of 1120 — 31% — are published and the rest are worked out.
Almost all of the derived figures come from holding output voltage constant rather than power, because that is what an amplifier actually does as the load changes. The exceptions are the 42 computed through a high output impedance, which use the voltage divider instead, and the 40 not yet traced to a document, which are not derived from anything. Below a maker's lowest published load the voltage is also clamped so the implied current never exceeds the highest current any published point implies, because current is the limit a real output hits first. The method is written up in full.
| Reason | Products |
|---|---|
| the product page or the whole product is gone | 33 |
| a power figure with no load attached to it | 31 |
| the maker publishes no headphone specification at all | 30 |
| only a voltage, measured with no load attached | 23 |
| the specification is published only as a picture | 21 |
| the product has no headphone output | 13 |
| other headphone numbers are published, but never a power | 11 |
| checked, but no power at a stated load could be found in a maker document | 10 |
| the model turned out not to exist | 9 |
| the maker’s own site could not be read at all | 6 |
| the maker never says which output the figure belongs to | 5 |
| the amplifier is built into an earphone cable, with no headphone jack | 5 |
| only one load, and too high to work down from | 4 |
| the maker's own numbers disagree with each other | 3 |
| the specification is assembled by a script, not published as text | 3 |
| other | 1 |
One group deserves to be stated precisely, because it is a fact about the search rather than an accusation: for 10 products, no power figure at a stated load could be found in any document the maker publishes. Some of those companies publish a detailed specification that simply has no power row in it, some put the table behind JavaScript, and some publish it only as a photograph of a table.
The groups after it are sharper, and these are counts of products rather than of companies. 30 products are listed with a headphone jack and no electrical description of it whatsoever — the jack appears in the connector list and nowhere else. 21 have a specification that exists only as an image, which is a decision somebody made rather than an accident — most often Topping (8), HiBy (4), xDuoo (3), Shanling (2). 31 carry a power figure with no load attached to it, which tells you nothing at all: a watt into 16 ohms and a watt into 600 ohms are different amplifiers. And 23 publish only a voltage measured with nothing plugged in, which cannot give a power at any load, because it says nothing about current.
Audient's interface spec sheets print three numbers on one line: a peak voltage, an RMS voltage, and a power. On the iD4 MkII at 30 ohms those are 2.59 V peak, 1.827 Vrms and 223 mW. But 1.827 Vrms into 30 ohms is 111 mW — half of what is printed. The same factor runs through all nine published points across the iD4 MkII, iD24 and EVO 4 — six of them land on 2.00 ± 0.02 and the other three sit 3–6% below it, which is what figures printed to two or three significant digits will do — and on the iD sheets the two voltage columns are internally consistent (the printed RMS is the printed peak divided by √2). The EVO sheets print only the peak, so the RMS used here for those three points is Vpk/√2 rather than a figure Audient printed.
Two ordinary explanations produce a factor of two here, and the arithmetic cannot tell them apart: squaring the peak voltage instead of the RMS one, or summing both channels into a single figure. Vpk²/R and 2·Vrms²/R are the same number, and Audient does not say which it means. Under either of those readings one channel driving one 30 ohm headphone gets 111 mW, not 223. This database uses the power that follows from Audient's own printed RMS voltage, keeps the printed voltages as published, and says so on every affected page. That correction covers the five interfaces whose figures were traced first: the iD4 MkII, iD14 MkII, iD24, EVO 4 and EVO 8. Six later Audient units — the iD48, iD44 MkII, EVO 16, Horizon 8P, ORIA and NERO — print the same shape of table, and their pages carry Audient's own printed figures with the discrepancy stated on the page rather than the numbers rewritten. Which of the two treatments is right depends on something Audient has not said.
The G111 Mk II datasheet prints load, dBu, volts and milliwatts in four columns. On three of its six rows the milliwatt column does not match the volt column: 300 ohms is printed as 1050 mW where 16.5 V implies 908 mW, 50 ohms as 2050 mW where 9.2 V implies 1693 mW, 32 ohms as 1400 mW where 6.2 V implies 1201 mW. The dBu column agrees with the volts on every row, so the volts are self-consistent and the milliwatts are the odd column out. A benign explanation exists — the power column may be quoted at a different distortion or duration condition than the voltage column — but the datasheet states only one condition for the whole table, and the same table appears in three separate official PDFs. This site records what Lake People printed and flags the gap rather than quietly picking a side.
Apogee gives “15 dBu / 240 mW @ 32 ohms”. Read as one statement it does not work: 15 dBu is 4.36 Vrms, which into 32 ohms would be 593 mW, while 240 mW into 32 ohms is 2.77 Vrms, or 11.1 dBu. Read as two statements it does: 15 dBu is the open-circuit ceiling, with no load attached, and 240 mW is what the output actually delivers into 32 ohms — 86.6 mA, an unremarkable current limit for a bus-powered box. The load qualifier sits at the end of the line and it is genuinely ambiguous which of the two numbers it governs. Both are recorded as printed, and only the 240 mW is treated as a power into a load.
Creative publishes something most makers do not: an output voltage and a power at every load, on both jacks of the Sound Blaster X5. Nine of the ten rows agree with themselves to the printed digit. The tenth does not: the balanced output is given as “5.86 Vrms (313 mW) @ 150Ω”, and 5.86 Vrms into 150Ω is 229 mW. The rows on either side of it imply about 8.28 Vrms, which would be 457 mW. Whatever happened there, the printed 313 mW is what this database stores, and the point of publishing both columns is that anyone can check — which is exactly why publishing both columns is the right thing to do.
The WiiM Ultra is rated at 150 mW into 32Ω and 6 mW into 300Ω. 150 mW into 32Ω is 2.19 Vrms; if the output can swing that, it would give about 16 mW into 300Ω, not 6. Six milliwatts into 300Ω implies 1.34 Vrms, and an amplifier that can only manage 1.34 V could not have produced the first figure either. Both are recorded as printed. Worth knowing before buying one for 300Ω headphones, because the honest reading is the pessimistic one.
Balanced outputs are usually sold as the powerful ones. SPL publishes the opposite for the Phonitor xe: 5 W into 250 ohms single-ended against 3.5 W balanced, and 2.7 W against 0.7 W at 32 ohms — along with a higher source impedance and a higher noise floor on the balanced jack. It is published clearly and it is worth reading before assuming the four-pin socket is the better one.
111 of the 280 sources here have no output impedance that can be traced to something the maker published, and 197 have no traceable noise figure. That count is deliberately strict: 10 early entries in this database do carry both figures, but they came from independent measurement rather than from any document, so they are counted here alongside the makers who published nothing. Their own pages say so.
It matters more than it sounds: output impedance is what decides whether a source will tilt the frequency response of a multi-driver in-ear, and it is a single number that costs nothing to measure.
Where it is published, the spread is enormous: from a claimed 0 ohms at one end to the Bottlehead Crack's 120 at the other. Merging's Anubis states 0.035 ohms. Focusrite's Scarlett Solo and 2i2 (4th gen) and three Audient interfaces state 50 ohms; the ART HeadAmp 4 Pro states 47. Either figure needs a 376–400 ohm headphone to satisfy the one-eighth rule — and only 6 of the 567 non-electrostatic models in this database are that high. None of those makers is hiding the figure; they publish it, which is more than most do. The rule, and what it does to your headphones, is here.
Reading a spec sheet carefully means reading which section a number sits in. On 29 of the 280 sources here an ohm figure does appear on the maker's page, and this database still records no headphone output impedance — because the figure describes something else. Each of those 29 product pages now names the figure, quotes it, and says what it actually belongs to, so the gap reads as a decision rather than an oversight.
15 of them print a load range: PreSonus's “Impedance Working Range: 32 Ω to 300 Ω”, M-Audio's “Impedance Range | 32 Ω - 600 Ω”, Fosi's “Headphone Impedance Match: 16-300 ohm”, McIntosh's four selectable taps. Those say which headphones to plug in. 7 print a real output impedance that belongs to a different socket — Steinberg's “Output Impedance 150 Ω” sits in the line output table, Universal Audio's 100 Ω under line outputs, Gustard's 200 Ω in the pre-amp block, Little Dot's 600 Ω against the RCA pre-out. Copying that last one across would have ranked a headphone amplifier against every headphone in this database on a number from its line stage.
Four are odder. Focusrite's Vocaster Two page has an “Output impedance” row inside its Headphone Outputs table, and the value printed in it is “20Hz - 20KHz ±0.5dB” — the frequency-response string from the row above. Universal Audio puts the same 100 Ω under line outputs in its specification article and under a headphone heading on the Volt 1 product page; both cannot be true of one output stage. Shanling gives the H2 “0.4 Ohm / 0.8 Ohm” without saying which jack gets which. And PreSonus's Quantum ES 2 carries a single “22 Ohms” under the label that holds a range everywhere else in its own catalogue.
One maker answers the question without a number at all: SMSL's C200 row reads “Output impedance: Line output 100 Ω / Headphone out Near 0 Ω”. That is a claim rather than a measurement and there is nothing to compute with, but it is not silence, and this site does not file it as silence. And on 2 products the only figure in circulation was published by somebody other than the maker — a magazine for the Woo Audio WA6, a dealer for the xDuoo TA-30. Neither is recorded as a specification.
Behringer's U-PHORIA datasheet earns the same treatment for a different field: it prints a maximum output level that belongs to the main outputs, and the word “Phones” appears only as a connector type.
A published power figure is only as good as the load printed next to it. If a spec sheet says “1 W output” with no impedance, it has told you nothing: 1 W into 16 ohms and 1 W into 600 ohms are different amplifiers. If it gives a voltage and no power, you have a ceiling rather than a rating, and the real figure into a low impedance may be far below it.
Every source page here says which of its numbers came from the manufacturer and which this site worked out, load by load. Browse them all, or download the lot with the provenance column intact. The products whose makers filled in all three figures — power at a stated load, output impedance and noise — are listed on their own page.
Reading every sheet turns up something the individual pages cannot show. In 15 cases a group of products publishes exactly the same figures for the headphone output — the same power at the same loads, and the same output impedance and noise wherever those are given at all. That is 36 of the 280 products in this database, a little over one in eight, whose headphone jack is indistinguishable on paper from something else here.
In 10 of the 15 groups the products do not cost within a hundred dollars of each other. The widest is the Sony NW-WM1AM2 at $1399 against the Sony NW-WM1ZM2 at $3700: $2301 between them, and not one decimal place of it visible in what the maker prints about the headphone output.
None of this says the dearer one is not worth the money. A headphone-output specification covers three numbers; it says nothing about converters, inputs, cases, batteries, volume controls or how any of it sounds, and those differences are real. What it says is where the spec sheet stops being the thing you are comparing, and for these 36 products it stops early. Each of their pages names the others.