Electronics
Valves or transistors? Where the “warm” sound comes from
In short: A valve amplifier really can sound different from a transistor one. There is no magic in it, though, just three things you can measure: more low-order distortion, mainly the second harmonic, soft clipping when overloaded, and above all a high output impedance, which forms a voltage divider with the speaker's impedance and shifts the frequency response by several decibels. Match the levels in a blind test, avoid clipping and use a speaker with a flat impedance, and the differences mostly disappear. A valve amp is neither better nor worse. It is a different tool: less power, more heat and upkeep, beautiful looks and a colouration some people love. If you want one, build it a speaker with high sensitivity and a flat impedance.
In this article:
- A short history
- How a valve amplifies and how a transistor does
- Reason one for the “warmth”: the second harmonic
- Reason two: soft clipping
- Reason three, and the biggest: output impedance
- The output transformer
- SET and push-pull
- Power and efficiency
- Heat, lifespan, cost and upkeep
- What blind tests say
- Valve preamp, transistor power amp
- A speaker that suits a valve amp
- Which one when
- Key takeaways
- Short glossary
A short history
In 1904 John Ambrose Fleming patented the vacuum diode: a glass bulb with a glowing filament and a metal plate. Current could flow through it in only one direction. Two years later Lee de Forest added a thin grid between the filament and the plate, and the triode was born. It was the first component that could amplify a weak signal. For the next half century valves ruled radios, telephone lines, cinemas, record players and the first computers.
In December 1947 John Bardeen, Walter Brattain and William Shockley demonstrated the first transistor at Bell Labs. They won the Nobel Prize for it in 1956. The transistor was small and cheap, needed no heating and barely wore out. In the 1960s and 70s it pushed valves out of most home electronics.
The first transistor amplifiers were far from perfect, though. Some had strong crossover distortion, a flaw at the point where the output stage hands over between the positive and negative halves of the wave. That is how “transistor sound” became an insult. Today's transistor amplifiers no longer have these flaws, but the reputation stuck. Since the 1980s and 90s valves have made a comeback, mainly in hi-fi and guitar amps. Japanese enthusiasts of simple amplifiers built around a single triode played a big part in it. More on those below.
How a valve amplifies and how a transistor does
Both do essentially the same job: a weak signal controls a large current from the power supply. Think of a water tap. Your hand moves only a little, with little force, yet the flow of water changes a lot. An amplifier is a tap that the music turns thousands of times a second.
The valve
- The cathode is heated, so electrons boil off it. That is where the heating and the orange glow come from.
- The anode sits at a high positive voltage, typically 200 to 450 V, and attracts the electrons.
- The grid between them carries a small negative voltage. A small change in that voltage makes a big change in the electron current reaching the anode.
So a valve works with high voltage and small current. A speaker wants the opposite: low voltage and large current. That is why a valve power amplifier almost always needs an output transformer to convert one into the other.
The transistor
A transistor is a sliver of silicon with three leads. In a bipolar transistor, a small current into the base controls a large current from collector to emitter. In a MOSFET, the voltage on the gate does the same. A transistor handles large currents at low voltage, so it can drive a speaker directly, with no transformer.
Modern amplifiers of both kinds use negative feedback: part of the output is fed back to the input, and the amplifier corrects its own errors. Transistor amplifiers use a lot of it, which is why their distortion is in the thousandths of a percent and their output resistance is almost zero. Many valve amplifiers, especially simple ones, use little or no feedback. That is where most of the differences in this article come from.
Reason one for the “warmth”: the second harmonic
Every amplifier distorts a little. To the tone you send in, it adds multiples of that tone, called harmonics. To a 440 Hz tone it adds 880 Hz (the second harmonic), 1,320 Hz (the third), and so on.
A triode's transfer curve is close to a parabola. A bend like that produces mainly the second harmonic, which sits exactly an octave higher. The octave is the sweetest interval in music, so the ear easily “hides” it inside the tone. The result is often described as a fuller, richer or warmer sound. Higher harmonics, such as the seventh or ninth, no longer fit the tone and sound harsh. Those come mostly from hard clipping and crossover distortion.
| Amplifier | Typical distortion (THD) at medium power | Dominant |
|---|---|---|
| Triode SET, no feedback | around 0.5 to 5 % | 2nd harmonic |
| Valve push-pull with feedback | around 0.1 to 1 % | 3rd harmonic |
| Good transistor amp (class AB or D) | below 0.01 to 0.05 % | varies, but very low |
| Bass driver, for comparison | around 0.3 to 3 % when playing loud, more in the bass | 2nd and 3rd harmonic |
The values are a rough guide. Exact figures depend on the specific circuit and power. The last row matters: the speaker itself distorts about as much as a SET. One percent of distortion dominated by the second harmonic is often hard to hear. Research by Earl Geddes and Lidia Lee showed that the THD figure is a very poor predictor of how objectionable distortion sounds. A few tenths of a percent of high harmonics bother the ear more than several percent of the second.
Reason two: soft clipping
Every amplifier has a ceiling. When the music goes past it, the peaks of the wave get “cut off”. A transistor amp with heavy feedback plays cleanly right up to the ceiling and then clips sharply, like driving a car into a wall. That creates lots of high harmonics, and the sound suddenly turns hard and harsh.
A valve approaches its ceiling gradually. Distortion rises slowly, like a car braking before the wall. It rounds off short peaks, such as a drum hit, and the ear finds that less disturbing. That is why an 8 W valve amp often sounds louder than you would expect from 8 W. Guitarists have even made this trait part of their sound.
For hi-fi the lesson is simple: when comparing amplifiers, neither may clip. Compare an overloaded transistor amp with an overloaded valve amp and the valve will win, but not because it is better. The transistor amp simply needed more power. More on power in How many watts do you really need.
Reason three, and the biggest: output impedance
This is the least known cause of the differences, and also the biggest. Every amplifier behaves as if it had a small resistor at its output, in series with the speaker. This is called the output impedance.
- Transistor amplifier: usually 0.01 to 0.1 Ω.
- Valve push-pull with feedback: usually 0.5 to 2 Ω.
- SET without feedback: often 1 to 4 Ω, sometimes even more.
Manufacturers often quote it as the damping factor: the speaker's nominal impedance divided by the output impedance. A transistor amp has a damping factor of 100 to 1,000, a SET into 8 Ω around 2 to 8.
A speaker, however, does not have a constant impedance. The bass driver has a peak at resonance, the crossover adds another hump near the crossover frequency, and in between the impedance drops almost to the resistance of the voice coil. The amplifier's output impedance and the speaker form a voltage divider, exactly the same one you know from the article on speaker cables. Where the speaker's impedance is high, it gets almost the full voltage. Where it is low, more of the voltage is lost at the amplifier's output.
The maths: how much the response changes
Voltage divider: voltage at the speaker = amplifier voltage × Z / (Z + Zout). In decibels: 20 × log(Z / (Z + Zout)). Z is the speaker's impedance at a given frequency, Zout the amplifier's output impedance.
Take a speaker whose impedance varies from 3.2 Ω to 30 Ω, and a SET with an output impedance of 3 Ω:
- at 30 Ω: 20 × log(30 / 33) = −0.8 dB,
- at 3.2 Ω: 20 × log(3.2 / 6.2) = −5.7 dB,
- difference: 4.9 dB.
The same calculation for other amplifiers:
| Output impedance | At 30 Ω | At 3.2 Ω | Response ripple |
|---|---|---|---|
| 0.05 Ω (transistor) | −0.01 dB | −0.13 dB | 0.12 dB |
| 1 Ω (valve with feedback) | −0.28 dB | −2.4 dB | 2.1 dB |
| 3 Ω (SET) | −0.8 dB | −5.7 dB | 4.9 dB |
A ripple of 2 to 5 dB over broad bands is clearly audible. For comparison, differences of around 0.5 dB across a wide band can already be picked out in blind tests. The change is not random: it follows the impedance curve. With our model speaker, the SET lifts the bass around resonance and the band around the crossover frequency, and dips the region around 300 Hz and the very top of the treble.
This is the “warm” sound in numbers: a slightly richer bass around resonance, a slightly recessed midrange and a rolled-off top end. On top of that, a higher output impedance damps the bass driver less well. Its total Q factor, Qts, goes up, and the bass becomes softer and longer. In a sealed box with a Qts of 0.7 that can mean a mild hump; in a bass reflex, bass that lingers longer.
Try it at home: put a 2 to 3 Ω power resistor, rated at least 10 W, between a transistor amp and the speaker. You will get a large part of the “valve” colouration without a valve. Then match the levels, or the comparison will not be fair.
The output transformer
A valve wants a load of several kilohms, while a speaker has 4 or 8 Ω. The transformer bridges the gap much like a gearbox in a car: high voltage and small current on one side, low voltage and large current on the other. The impedance ratio is the square of the turns ratio, so going from 3,000 Ω to 8 Ω takes a turns ratio of about 19 : 1.
The output transformer is the heart of a valve amplifier and often its most expensive part. It brings its own limits:
- Bass: at low frequencies the core can saturate and distortion rises. A cheap transformer runs out of breath here sooner.
- Treble: leakage inductance and winding capacitance limit the top of the band. A good transformer reaches 20 kHz and beyond, a poor one does not.
- Taps: the secondary often has outputs for 4, 8 and 16 Ω. Connect the speaker to the tap that matches its nominal impedance. A lower tap also means a lower output impedance.
- Weight and cost: a transformer for 10 W with full bass weighs several kilograms.
There are also amplifiers without an output transformer, called OTL. They need many valves in parallel and work best with high-impedance speakers. They are rare.
SET and push-pull
SET: one valve, pure simplicity
A SET (short for single-ended triode) has a single output triode per channel that amplifies the whole wave. It runs in class A, meaning it conducts current all the time. Typical valves and their output:
- 45: around 1.5 to 2 W,
- 2A3: around 3 to 3.5 W,
- 300B: around 7 to 8 W, the legend from the Western Electric workshops,
- 211 or 845: 15 to 20 W, but with an anode voltage of around 1,000 V.
A SET has the simplest circuit, a dominant second harmonic and the highest output impedance. A constant current flows through its transformer, so the core needs an air gap, which makes it large and expensive. Power is low and distortion climbs quickly with volume.
Push-pull: two valves working in opposition
In a push-pull circuit, two output valves take turns: one “pushes”, the other “pulls”. Their even harmonics cancel each other out in the transformer, leaving mainly a weaker third. The constant current also cancels in the transformer, so the core can be smaller.
Common valves are the EL34, KT88, 6L6 and EL84. A pair of EL84s gives around 15 to 17 W, a pair of EL34s around 30 to 40 W, and a pair of KT88s around 50 W or more. A push-pull amp with feedback measures far better than a SET and copes better with ordinary speakers. Classic examples are the Williamson amplifier of 1947 and the Mullard 5-20 from the 1950s.
| SET | Push-pull | |
|---|---|---|
| Power | 1 to 8 W, exceptionally 20 W | 15 to 100 W |
| Distortion | higher, mainly 2nd harmonic | lower, mainly 3rd harmonic |
| Output impedance | high, 1 to 4 Ω | lower, 0.5 to 2 Ω |
| Speaker requirements | high sensitivity, flat impedance | ordinary speaker with a sensible impedance |
| Complexity | simple, expensive transformer | more parts, matched valves |
Power and efficiency
Efficiency tells you how much of the power drawn from the wall socket reaches the speaker. The rest is heat.
- SET in class A: the valve itself turns around 20 to 30 % of what it receives into output power. Count the heaters as well and the total is even lower. A stereo 300B amp draws around 100 to 150 W from the wall, even when it is playing quietly, and delivers 2 × 8 W.
- Valve push-pull in class AB: somewhat better, around 30 to 50 % excluding the heaters.
- Transistor in class AB: around 50 to 70 % at full power, with a small draw at idle.
- Class D: 85 to 95 %, it barely gets warm.
More on classes in the article Amplifier classes.
How many decibels 8 W gets you
Maximum loudness at 1 m ≈ sensitivity (dB/1 W/1 m) + 10 × log(power in W). At 3 m from the speaker, a typical room takes away roughly 5 to 7 dB.
| Speaker sensitivity | 2 W | 8 W | 100 W |
|---|---|---|---|
| 84 dB (small bookshelf) | 87 dB | 93 dB | 104 dB |
| 90 dB | 93 dB | 99 dB | 110 dB |
| 96 dB (full-range, horn) | 99 dB | 105 dB | 116 dB |
Music has peaks 10 to 20 dB louder than its average. With an 84 dB speaker and 8 W, the average level at the sofa comes out at only a little over 70 dB, and the amplifier will already be clipping the peaks. With a 96 dB speaker you get 12 dB more, which is the same as 16 times the power. That is why SETs have always gone hand in hand with sensitive speakers: full-range drivers, horns and big two-ways. Have a look at the articles Famous designs and the Needle build diary. The small 10 cm full-range driver in the Needle has rather average sensitivity, though, so I would not pair it with a 2 W SET. With an 8 W 300B in a small room it works.
Heat, lifespan, cost and upkeep
Careful: a valve amplifier runs at 300 to 500 V, and big triodes at around 1,000 V. The filter capacitors hold that charge long after switch-off. Do not reach into an amplifier that is on or has just been switched off unless you know what you are doing.
- Heat: output valves reach 100 °C or more on the glass. The amplifier needs free space above it, does not belong in a closed cabinet, and you need to keep an eye on children and cats.
- Lifespan: output valves last in the order of a few thousand hours, depending on type and load. Small preamp valves usually last much longer. A worn valve loses power, gets noisy or turns microphonic, meaning it rings when tapped.
- Bias adjustment: many fixed-bias amplifiers need adjusting after you replace the output valves, and push-pull valves should be bought in matched pairs. Auto-bias amplifiers take care of this themselves.
- Capacitors: the electrolytic capacitors in the power supply age after 15 to 25 years, as in any electronics, only at a higher voltage.
- Cost: a good output transformer, mains transformer and high-voltage supply are expensive. For the same money you can get a transistor amplifier with many times the power. A set of output valves costs roughly anywhere from a few tens to a few hundred euros, and more for rare types.
A transistor amplifier, on the other hand, runs for years without maintenance, and class D does not even get warm. That is an argument for convenience, not for sound.
What blind tests say
A fair amplifier comparison has to meet three conditions. Without them the result is worthless, however convincing it sounds:
- Matched levels to within a tenth of a decibel, measured with a voltmeter at the speaker terminals. The louder amplifier almost always “sounds better”.
- No clipping. The volume must stay low enough that even the weaker amplifier never hits its ceiling.
- Blind, with quick switching, for example using the ABX method. How to do it is explained in Hi-fi and high-end. You can try it yourself in the ABX blind test.
When these conditions are met and the amplifiers have low output impedance and low distortion, listeners usually cannot tell them apart in blind tests. This result has been repeated since the 1970s. A difference reliably shows up when the frequency response differs, for example precisely because of output impedance, or when one amplifier clips.
That means two things. First, a SET with a 3 Ω output on a speaker with a wildly swinging impedance can be told apart, and quite rightly so. It is not imagination, it is physics. Second, that difference is colouration, not higher fidelity. A valve does not add information missing from the recording. It changes the response and adds harmonics. If you like that, it is perfectly fine; just call it by its real name.
An interesting case is Bob Carver's experiment from the mid-1980s. He modified a transistor amplifier to mimic the transfer function of a specific valve amplifier, including its output impedance. Editors of Stereophile magazine, who did not believe it could be done, then could not reliably tell the two apart by listening. It shows that “valve sound” can be described by measurements and imitated.
Valve preamp, transistor power amp
A popular compromise is a valve preamplifier followed by a transistor power amplifier. What do you gain and what do you lose?
- Output impedance stops being a problem. The transistor power amp holds the speaker firmly, so the response stays flat and the bass is well damped.
- There is plenty of power, even for ordinary speakers with a sensitivity of 85 dB.
- The colouration is weaker. A preamp works at low voltages, far from its ceiling, so its distortion is typically a few tenths of a percent or less. A good valve preamp measures almost neutral. If you are after strong “warmth”, you will not get it here.
- Mind the connection: a valve preamp can have an output impedance of hundreds of ohms up to several kilohms. The power amp's input should be at least ten times higher, ideally 47 kΩ or more, and the cable should be short.
A speaker that suits a valve amp
If you are building a speaker for a valve amplifier, most of the result is in your hands. That goes double for a SET.
1. High sensitivity
Aim for at least 90 dB/1 W/1 m, and for a SET preferably 93 dB or more. A larger bass driver with a light cone and a strong magnet helps, as does a full-range driver or a horn for the treble. The price is usually less deep bass for a given box size; that is physics you cannot get around. Watch out when comparing: sensitivity quoted at 2.83 V for a 4 Ω speaker is actually measured at 2 W, not 1 W, so it looks 3 dB “prettier”.
2. Flat impedance
The less the impedance jumps around, the less an amplifier with a high output impedance changes the response. You have three tools:
- Zobel network: a resistor and capacitor in series, connected in parallel with the bass driver. It flattens the impedance rise caused by the voice coil's inductance, so the crossover sees the resistance it was calculated for.
- LCR network for the resonance: a series circuit of an inductor, capacitor and resistor in parallel with the driver, tuned to its resonance. It tames the impedance peak, for example on a tweeter or a bass driver in a sealed box.
- A crossover without deep dips: do not let the total impedance fall below about 0.8 times its nominal value. For 8 Ω, that means not below 6 to 6.5 Ω.
Loudsmith has a Zobel network built right into the crossover and shows the impedance curve of the whole speaker. When designing for a valve amp, watch both curves: response and impedance. If the impedance only spans, say, 6 to 10 Ω, even a SET with a 3 Ω output ripples the response by only about 1.2 dB instead of nearly 5 dB.
3. Higher nominal impedance
An 8 Ω speaker is a kinder load for a valve amp than a 4 Ω one. The ratio of output impedance to speaker impedance is more favourable, and the transformers are often optimised for 8 Ω.
4. Design passive EQ with the amplifier in mind
Series correction networks, like the one in the Needle (2.7 Ω ∥ 0.18 mH ∥ 22 µF in series with the driver), tame a hump by raising the impedance in that band. How well they work depends on what sits in front of them. With a 4 Ω driver and a transistor amp, this network pulls the hump around 2.5 kHz down by about 4.5 dB. With an amplifier that has a 2 Ω output, only by about 3.2 dB. If you know you will be playing from a valve amp, include its output impedance in the design as a series resistance.
If you care less about the valve itself and more about a speaker that sounds good with anything, see also active and passive speakers. In an active speaker each driver connects directly to its own amplifier, and output impedance plays no role.
Which one when
| Situation | Recommendation |
|---|---|
| You want neutral sound, lots of power, zero maintenance | transistor amplifier, class AB or D |
| Ordinary speakers with 84 to 88 dB sensitivity | transistor; a valve push-pull of at least 30 W if you want valves |
| Sensitive speakers above 93 dB, small room, quieter listening | a SET is a sensible choice if you like its colouration |
| You want the look and atmosphere of valves, but neutral bass | valve preamp and transistor power amp |
| You are building a speaker for a valve amp | 8 Ω, high sensitivity, Zobel and LCR, flat impedance |
| You just want to hear how a “valve” would change your speakers | a 2 to 3 Ω resistor between the transistor amp and the speaker, with matched levels |
A valve amplifier is like a mechanical watch. It keeps time less accurately than a cheap digital one, needs servicing and costs more. Plenty of people love it anyway, for the craftsmanship, the looks and the feel. That is a perfectly legitimate reason. Just do not believe it tells a “truer” time.
Key takeaways
- The “warm” valve sound is measurable. It comes from the second harmonic, soft clipping and, above all, high output impedance.
- Output impedance shapes the response according to the speaker. A SET with a 3 Ω output can ripple the response of an ordinary speaker by up to 5 dB.
- In a fair comparison the differences mostly disappear. Matched levels, no clipping and a speaker with a flat impedance.
- A valve amp is colouration, not higher fidelity. If you like it, that is fine; just know what you are paying for.
- For a valve amp, build a sensitive speaker with a flat impedance. That is where you get the best of both worlds.
That last point is where you make the call, not the amplifier. In Loudsmith you can see both the sensitivity and the impedance curve of your design, and flatten it with a Zobel network before you buy a single part.
Short glossary
- Triode: a valve with a cathode, grid and anode. The simplest amplifying valve.
- Pentode, beam tetrode: valves with extra grids, such as the EL34 or KT88. They offer more power and efficiency.
- Harmonics: multiples of the fundamental tone that distortion adds.
- THD: total harmonic distortion, in percent.
- Feedback: part of the output is fed back to the input, and the amplifier corrects its own errors.
- Output impedance: the amplifier's internal resistance in series with the speaker.
- Damping factor: the speaker's nominal impedance divided by the amplifier's output impedance.
- Output transformer: converts the valve's high voltage into the low voltage and large current a speaker needs.
- SET: an amplifier with a single output triode per channel, class A.
- Push-pull: two output valves working in opposition; even harmonics cancel out.
- OTL: a valve amplifier without an output transformer.
- Bias: the valve's resting operating point, around which the signal “swings” up and down.
- Zobel network: an R and a C in parallel with the driver that flatten the impedance rise in the treble.
- LCR network: a circuit that tames the impedance peak at a driver's resonance.
Sources: R. O. Hamm, Tubes Versus Transistors: Is There an Audible Difference? (JAES 1973); E. Geddes and L. Lee, Auditory Perception of Nonlinear Distortion (AES Convention 2003); D. T. N. Williamson, Design for a High-Quality Amplifier (Wireless World 1947); M. Jones, Valve Amplifiers (Newnes); D. Self, Audio Power Amplifier Design (Focal Press); F. E. Toole, Sound Reproduction: The Acoustics and Psychoacoustics of Loudspeakers and Rooms (Focal Press); I. G. Masters and D. L. Clark, Do All Amplifiers Sound the Same? (Stereo Review 1987); datasheets for the 45, 2A3, 300B, EL34, EL84 and KT88 valves. The voltage divider and correction network calculations and the modelled impedance in the charts are my own.
Free tools: Room planner · Speaker measurement · Blind ABX test