Why 12AX7, 12AT7, 12AU7, 12BH7 and 12BZ7 Tubes Sound Different
Why can swapping a small tube change an amplifier’s gain, brightness and distortion? Explore five miniature twin triodes through interactive waveforms, harmonics, load lines and supply behavior, with original manufacturer sources and the modeling assumptions explained.
12AX7, 12AT7, 12AU7A, 12BH7A and 12BZ7, compared through gain, loading, harmonics, headroom and supply behavior.
Substituting a tube can change a circuit in several audible ways. The useful dividing lines are gain, current, bias and capacitance. The first middle letter of the type number does not define a tonal family.
The tube names hide a wide electrical spread.
Evidence: original manufacturer sheetsThe table uses a stated 250 V plate-to-cathode characterization point for each triode section. The grid bias and plate current differ between types. These are different test conditions, not five interchangeable recipes for a preamp stage. GE's 12AT7 point uses a 200 Ω cathode resistor; approximately −2 V is inferred from 10 mA × 200 Ω.
| Type / source | μ | gm mS | rp kΩ | Ip mA | Vgk V | Cgp / Cin pF | Heater A at 12.6 V | Plate limit W / section |
|---|
All five are twin triodes with 9A basing. Capacitances are the cited unshielded values; “Cin” is grid-to-cathode/heater input capacitance, before Miller multiplication. BH7A DC data are RCA, with its capacitances and heater data checked against Tung-Sol. Ratings use the cited editions; design-center and design-maximum conventions differ. Tube ratings from a different edition or variant must be checked separately. [1–6]
Gain potential versus current sensitivity
gm measures current change per grid volt at fixed plate voltage. It is not a complete measure of load-driving ability.
Plate resistance is a second, separate axis
Lower rp generally helps a common-cathode plate node act as a voltage source. A cathode follower is governed more directly by gm and its load.
12AX7 and 12BZ7: same nominal μ, different rp, gm and capacitances. 12AU7A and 12BH7A: similar μ, different drive and thermal allowances. 12AT7: an A-letter device with the highest cited triode gm in this comparison.
The 12BY7A is a pentode with a screen grid; the 12BA7 is a heptode converter. They are retained as topology examples, but are excluded from the dual-triode simulations. Conversion transconductance is not interchangeable with ordinary triode gm. [7] [8]
Watch the sine bend. Then look at what it adds.
Evidence: illustrative nonlinear model, recomputed locallyA common-cathode stage inverts the signal. Current rises on a positive grid excursion, the plate resistor drops more voltage, and the plate voltage falls. Unequal curvature on the two halves of a cycle produces harmonics. A waveform can look nearly sinusoidal while its residual and spectrum reveal a difference.
The model is anchored to each tube's published current, μ and rp at one operating point. It is adequate for explaining the relationships; it has not been fitted to the full manufacturer plate curves. Its relative distortion ranking is therefore conditional. Strong-overdrive tone, grid conduction, blocking, thermal memory and noise are outside these calculations. [9] [10]
All circuits: 300 V supply; 1 MΩ output load coupled for AC; cathode-biased DC; 1 kHz sine. Rebiasing uses plate targets 180 V for AX7/BZ7 and 170 V for AT7/AU7A/BH7A, with tube-specific plate resistors. The common circuit solves a new DC point for each tube. “Held at Q” freezes cathode voltage for the AC calculation; rectified cathode rebiasing is omitted.
Actual AC plate output
Shared voltage scale, with DC removed. Compare height for gain, curvature for distortion. The two cycles occupy 2 ms at 1 kHz.
Shape after matching the fundamental
Divided by each output's fundamental amplitude. The dashed trace is the ideal inverted sine. This display matching does not change the electrical input.
Harmonics generated by the stage
H2 to H7 relative to each tube's fundamental (dBc). −40 dBc means 1% amplitude; −60 dBc means 0.1%. Display floor: −100 dBc. Total reported THD includes all sampled residual energy.
Distortion residual, enlarged
DC and the best-fit fundamental removed. A twice-per-cycle pattern is H2. The voltage scale is much smaller than in the actual-output plot.
| Current conditions | Input Vpk | Fundamental Vpk | THD % | Ip,Q mA | Plate,Q V to ground | Cathode,Q V | Ra / Rk kΩ |
|---|
Same input compares input sensitivity. Same output removes the gross difference in voltage amplification. Same fraction of bias compares relative grid excursion; it does not guarantee equal proximity to every clipping mechanism.
How distortion grows across all five types
Left: equal input voltage, with each trace ending at 90% of its own bias magnitude. Right: equal fraction of bias magnitude. Both use the currently selected circuit and cathode behavior. This is a model comparison, not a universal distortion specification.
Input swing and output swing answer different questions
The bar is the modeled quiescent cathode voltage with a DC-grounded grid: the input needed to reach Vgk = 0 if the cathode is held still. It is an approximate grid boundary, not “clean headroom.” With an unbypassed cathode, the cathode follows signal current and the static bar is not an exact threshold.
A lower-gain stage can accept a larger input before producing a given plate swing. That does not promise a larger maximum clean output. Supply voltage, bias location, loading and the low-voltage end of the plate curves limit the output separately. More gain can also overdrive the next stage earlier even when this stage itself remains reasonably linear.
The load line selects the part of the curve you hear.
Evidence: modeled plate curves and solved circuitModeled plate characteristics and load line
Transfer around the chosen operating point
The falling line expresses the DC resistor constraint. The dot is the quiescent operating point, Q. The AC load also includes the 1 MΩ following stage. With the cathode unbypassed, it moves with current and the transfer changes through local feedback. The tube never travels arbitrarily around a datasheet graph.
Vpk = Vplate − Vcathode; Vgk = Vgrid − Vcathode
|Av| ≈ μ(Ra ∥ RL) / [rp + (Ra ∥ RL)] for a bypassed cathode
Keeping all resistors unchanged during a substitution changes Q. Rebiasing to the same plate target is a different experiment. Notice the modeled AU7A and BH7A plate voltages fall substantially in the common 100 kΩ circuit. A type's published “typical 10 mA” point does not force 10 mA through a resistor that cannot supply it.
Unbypassed cathode resistance feeds some of the current change back into Vgk, reducing gain and usually reducing low-level distortion. A real bypass capacitor makes that feedback frequency dependent. At overload it can also produce time-dependent bias movement; this static comparison does not simulate that recovery.
More gm can still mean more treble loading.
Evidence: calculated one-pole approximationThe plate moves opposite to the grid in a voltage amplifier. The grid-to-plate capacitor therefore demands more charge from the source than its physical value alone suggests. This is the Miller effect. Larger Cgp and larger gain both increase the effective input capacitance.
fc ≈ 1 / [2π(Rsource ∥ Rgrid)Ceff]
Uses all five current DC points, an ideal AC-bypassed cathode, 1 MΩ grid leak and 1 MΩ output load. Source resistance is a pure resistor. The plots are normalized to each stage's own low-frequency gain.
Estimated high-frequency attenuation
Effective input capacitance
| Type | Gain used V/V | Ceff pF | Estimated pole kHz | At 20 kHz dB relative |
|---|
The BZ7's larger capacitances can create more high-frequency loading than AX7's behind a high-impedance source. Drag toward 1 kΩ and that difference shrinks across the audio band. This provides a credible route to a darker or less open sound in a particular circuit; it does not establish a universal subjective description.
A passive pickup has inductance, resistance, pots, cable capacitance and a resonance. It is not represented by this resistor-only model. Output-node poles, frequency-dependent gain and stray capacitances are omitted. The calculated fc is a one-pole estimate, not a complete amplifier bandwidth measurement. Cgp/Cin inputs are from [1–6]; dynamic modeling context: [10].
Sag needs a supply and a change in current.
Evidence: exact RC response to an assumed current stepThe voltage drop caused by standing current is a DC operating condition. Sag is a change over time. A supply resistor and reservoir capacitor let a current increase pull the node down, then recover when that increase ends. The step below is deliberately assumed; it is not derived from the sine-wave tube simulation.
Assumed current demand
Supply-node movement and recovery
After the step: ΔVB+(t) = ΔVB+(tend) exp[−(t − tend) / τ]
τ = Rfeed C; long-duration droop magnitude = ΔI Rfeed
A class-A preamp draws standing current even at silence. A signal does not automatically add a current step equal to its whole bias current. Average current change depends on curvature, bias, clipping and the connected stages. A calculation of DC voltage loss alone cannot establish a tube's musical sag.
Two other recovery mechanisms can sound similar: the cathode bypass capacitor can change bias, and grid current can charge a coupling capacitor, temporarily shifting the grid toward cutoff. The latter is blocking distortion. Those need multi-cycle dynamic models and are not included in the static THD plots. [9]
Follow the mechanism to the audible result.
→ earlier drive of following stages.
→ H2, H3 and changing overload texture.
→ treble attenuation or changed pickup resonance.
→ load compression and possible recovery effects.
| Situation | What to compare | Useful inference |
|---|---|---|
| First guitar-input stage | Gain, source impedance, Cgp, noise and microphonics | AX7/BZ7 can differ in treble loading as well as gain. Pickup and wiring details matter. |
| Plate-output driver | Actual bias current, rp ∥ Ra, and capacitive load | AU7A/BH7A may give a stronger voltage source if the circuit permits enough current. |
| Cathode follower | gm at the actual Q point, current reserve and load | AT7's higher cited gm matters. BH7A's dissipation allowance does not alone make it the best follower. |
| Phase splitter | Topology, balance, tail impedance, drive current and feedback | A single-ended H2 comparison does not predict a balanced phase splitter or push-pull amplifier. |
| Overdrive stage | Both input and output level, bias, coupling and grid current | Match output level before judging intrinsic texture. Strong clipping requires a fuller model. |
Current drive, slew and power-supply rejection
Charging a capacitance requires current: I = C dV/dt. A 20 kHz sine at 50 Vpk needs a maximum slope of about 6.3 V/µs. This is a requirement on the complete stage, not a tube's intrinsic slew-rate specification. At a common-cathode plate, the tube pulls down while the plate resistor largely supplies the upward charging current. More tube current alone does not remove that asymmetry.
With grid and cathode AC-grounded, a simplified plate-ripple divider is rp/(Ra + rp), or (rp ∥ RL)/[Ra + (rp ∥ RL)] with an AC output load. This describes supply-to-plate feedthrough under those assumptions. It is not the full amplifier's PSRR or hum performance; decoupling, cathode feedback and heater coupling can change the result.
Noise and microphonics need comparable tests
The selected sheets do not provide a matched noise test for all five types. There is no supported A-versus-B noise ranking. Microphonics depend on mechanical construction, sample condition, mounting and gain. RCA specifically describes the AU7A for applications critical to microphonics, which illustrates why a revision or specimen can matter within one type. [3]
For an auditionable model of hardware you plan to build, use the actual schematic and supply. Fit the plate and grid current curves over the intended range, include capacitances and coupling networks, and check the result against hardware. The demonstrations here explain mechanisms; they do not validate a finished audio-device design.
The circuit is the key: these tubes can sound different because they make the same circuit operate differently. Harmonic structure, input sensitivity, output loading, Miller capacitance and supply behavior are plausible causes. A type number alone does not tell us which cause dominated a particular listening experience.
Sources, equations and limits are visible.
Manufacturer values below use the original sheet scans, including their operating conditions and shielded versus unshielded capacitances. Numerical plots are circuit calculations and illustrative simulations, with their assumptions disclosed.
- GE 12AX7, ET-T509B, June 1953. Page 1: DC parameters, capacitances, heater, basing and ratings.
- GE 12AT7, ET-T1440, February 1957. Pages 1–2: capacitances, heater, ratings and 250 V / 200 Ω characterization.
- RCA 12AU7A, July 1961. Pages 1–2: DC parameters, unshielded capacitances and 2.75 W per-plate design-maximum rating.
- RCA 12BH7-A, March 1955. Page 2: 250 V characteristics and 3.5 W plate dissipation. Use the class-A limits rather than pulse ratings for audio design.
- Tung-Sol 12BH7A, July 1962. Pages 1–2: shield conditions, capacitances, heater and separate class-A / deflection ratings.
- Sylvania 12BZ7. Page 1: 250 V characteristics, unshielded capacitances, heater and 1.5 W rating.
- GE 12BY7-A, ET-T943, June 1955. Pages 1–2: pentode structure, pinout and distinct screen requirements.
- Tung-Sol 12BA7, October 1948. Heptode/pentagrid converter structure and pinout.
- Norman Koren, Improved vacuum tube models for SPICE. Explains the shortcomings of the older power-law model and the need to model coupling-capacitor charging during grid conduction. This report uses its own transparent illustrative fit, not Koren's fitted model library.
- Cohen and Hélie, Simulation of a guitar amplifier stage for several triode models, AES 127, 2009. Static versus dynamic stage simulation and parasitic-capacitance effects.
Exact model, calibration and numerical method
Units in the solver are volts, amperes, ohms and farads. Each tube uses a shifted three-halves-power illustrative current law:
S0 = (3/2) I0 rp0
k = I0 / S0^(3/2)
E0 = S0 − 250 − μVgk,0
This exactly reproduces the source current and local rp at the anchor. Its gm there is μ/rp, which agrees with the rounded source gm within about 2%. It does not establish the full transfer characteristic. In particular, it has inadequate low-plate-voltage and deep-cutoff physics. The modeled plate curves are labeled accordingly.
DC cathode bias is solved using Vpk = B+ − Ip(Ra + Rk) and Vgk = −IpRk. In the rebiasing mode Rk is calculated to place Q at the stated plate-to-ground target. For each AC sample, the resistor and tube-current equations are solved together, including the 1 MΩ AC output load. Grid current is omitted. The bypass option holds the cathode at its DC voltage; the unbypassed option solves cathode movement.
Waveforms use 512 equally spaced samples over one periodic cycle. The fundamental is obtained by sine/cosine projection. THD = RMS of the residual after removing DC and fundamental, divided by fundamental RMS. This includes all represented harmonic energy; there is no added noise. THD sweeps use 256 samples per cycle. Displays show two repeated cycles for clarity. No external downloads, fonts or chart libraries are needed to view this file.
The drive ceiling is 90% of the modeled quiescent bias magnitude. In the same-output mode, the input is solved independently for each selected tube; any ceiling is reported. Harmonic bars below −100 dBc are shown at the display floor, which is not a modeled noise floor.
Miller plots use small-signal gain obtained by a finite-difference derivative at Q. They always assume an AC-bypassed cathode and a simple 1 MΩ grid leak. The RC supply plot solves only an assumed extra-current step lasting 300 ms. It is independent of the static tube model.
What remains uncertain
The exact sound of your tubes in your circuit is unresolved without the schematic, measured supply and bias conditions, source/load impedances, and preferably recordings matched in level. Individual tube variation, frequency-dependent cathode feedback, positive-grid behavior, blocking, output capacitance, pickup resonance, hum and mechanical noise are not predicted by these static plots. No unsupported listening scores, “warmth” rankings or full-clipping thresholds are assigned.