Related article: Dedicated to those who have never heard a good transformer.

In this article, I want to bring some clarity—without delving into complex mathematical formulas that might be inaccessible to those approaching the topic for the first time—on what we might call the “specificity” of the output transformer. My goal is to offer a personal perspective on this theme, sharing reflections I’ve developed over time, and at the same time warn readers about certain conceptual distortions that unfortunately clutter the web. It’s not uncommon to run into individuals who, driven by frustration or personal rivalries, jump into technical debates only to discredit, confuse, or vent. I prefer to counter these dynamics with a constructive approach grounded in direct experience, aiming to offer useful content to those who genuinely want to understand, without prejudice or polemics.

Should every tube have its own transformer?

To tackle this topic, let’s start with a few concrete examples, considering some common tubes. We begin with the EL34, a widely used pentode. This tube has an internal resistance of about 15k? and a maximum plate dissipation of 25 watts. In a single-ended configuration, it is usually loaded with 2k? with an anode current around 100 mA. Under these conditions, the theoretical power one could obtain is about 11 watts. I deliberately use the term theoretical because, in practice, you must always account for losses introduced by the output transformer, which inevitably steals part of the available energy.

We can also look at the EL34 operating in push-pull, where the datasheet itself proposes several AB-class operating solutions. For example, three different configurations are listed, each associated with a different primary impedance for the transformer: 6600 ?, 3400 ?, and 2800 ?. These configurations differ not only in transformer impedance, but also in the type of bias used, the output power, and the resulting harmonic distortion.

Depending on the chosen operating point, you can obtain very different results: output power can vary from a minimum of about 20 watts up to a maximum of 55 watts. It is therefore evident how the transformer must be selected precisely for the intended application, taking into account the electrical characteristics of the tube as well as the goals in terms of performance and sound quality.

Another interesting possibility, often overlooked at the time due to the race for maximum available power, is class-A push-pull operation. In this configuration, it is perfectly reasonable to see two EL34s working on a transformer with a 4000 ? primary impedance. While giving up some of the power achievable in class AB, this choice can offer advantages in linearity, frequency response, and musicality—traits far more appreciated today than in the past.

Moving on to another very common tube, let’s analyze the 6V6. This tube has an internal resistance of about 50k? and, in a single-ended configuration, a recommended load of 5000 ?. Under these conditions, the theoretical power is about 4.5 watts. In this case too, the datasheet lists some push-pull configurations with 8k? and 10k? primaries that allow power up to about 14 watts, depending on the chosen bias and operating conditions. As with the EL34, the 6V6 shows how the relationship among tube, transformer impedance, and operating point is crucial in determining the amplifier’s overall behavior.

Let’s continue with the 6L6, another classic and much-appreciated tube. Its internal resistance is around 22.5k?. In a single-ended configuration, the typical recommended load is about 2500 ?, from which you get a theoretical power around 6.5 watts.

As for push-pull operation, to avoid weighing down the reading further, I refer you to the data shown in the image below. As always, just click the image to enlarge and conveniently consult all the technical information on the various operating configurations of the 6L6.

Now let’s look at the 2A3, a directly heated triode beloved by audiophiles for its timbral qualities. Its internal resistance is about 800 ? and, in a single-ended configuration, it can deliver a theoretical power of about 3.5 watts when loaded with 2500 ?. Interestingly, this is the same recommended impedance for the 6L6 in single-ended. But a question naturally arises: can a transformer designed for the 2A3 also be used for the 6L6? The answer is not so straightforward, and later in the article we’ll see if—and under what conditions—this may be true or not.

Now let’s move to the KT88, a powerful and versatile tube widely used in power amps. Its internal resistance is about 12k?. The official Genalex datasheet does not foresee using the KT88 in a single-ended configuration, although, in theory, it could deliver up to 17 watts. However, in real applications, the useful power settles around 10 watts with a 6k? transformer, before distortion becomes significant.

Regarding push-pull operation, datasheets list numerous configurations in AB1 and AB2. For brevity, I’ll mention only three, employing primary loads of 5k?, 6k?, and 4.5k?. Under these conditions, the theoretical output power can reach up to 100 watts, depending on the operating point and type of bias used.

So far it should be quite clear that all the tubes analyzed are profoundly different from each other: each requires a specific load impedance, has its own distinct internal resistance, and operates under unique electrical parameters.

An important point to stress is that the transformer’s primary impedance never equals the tube’s internal resistance. I highlight this because one occasionally comes across rough tests where a simple series resistor (equal to the primary impedance value) is used to evaluate an output transformer. This approach leads to completely misleading, technically baseless results and provides no reliable information on the transformer’s real performance.

Moreover, every tube has a different maximum dissipation, which translates into varying operating points in both voltage and current, with resulting differences in efficiency and actual deliverable power. All these factors make each tube-transformer pairing a case in itself, requiring care, knowledge, and experience in design.

Let’s return to the case mentioned earlier, that of the 6L6 and the 2A3, two tubes seemingly very different from each other, yet for which the datasheet recommends—purely by coincidence—the same primary impedance: 2500 ?. As we have seen, the 6L6 can deliver a theoretical power of about 6.5 watts, while the 2A3 tops out around 3.5 watts. At first glance, one might think: “If I build a 2500 ? transformer capable of handling the 6.5 watts of the 6L6, it will be perfect for the 2A3, which runs at lower power.” But is it really that simple?

To answer, let’s return to a fundamental detail already mentioned: the internal resistance (Ri) of the two tubes. For the 2A3, Ri is about 800 ?, while for the 6L6 it’s a hefty 22,500 ?—a huge difference. And this is precisely where a parameter often neglected—or worse, weaponized by those spreading incomplete or distorted information—comes into play: the transformer’s primary inductance.

Have you ever heard it said that “a good output transformer must have at least 100 Henry”? Or that “if it has less, it’s not suitable for audio”? Or that certain manufacturers make transformers with 200, 300, or even 700 Henry of primary inductance—as if it were a contest over who has the highest figure?

These claims must be contextualized and, above all, critically analyzed, because the required inductance depends strictly on the type of tube used, its internal resistance, the desired cutoff frequency, and other design parameters. In the next paragraph we’ll see why this all makes sense—and why it doesn’t make sense to use the same transformer for two such different tubes, even if they share the same primary impedance value.

Let’s imagine a theoretical circuit consisting of a signal generator, a resistor representing the tube’s internal resistance (Ri), and an inductor corresponding to the output transformer’s primary inductance. Those who want to slog through formulas can consult Wikipedia.

SPICE can help simplify understanding this phenomenon. Let’s imagine R1 represents the internal resistance (Ri) of our 2A3 and L1 is the primary inductance of our transformer, with a theoretical value of 20 H. Now, let’s observe the passband:

In the graph, we can see that at 1 kHz the signal level is 6 dB. If we shift the analysis to about 12 Hz, we see the signal drop to 5 dB (equivalent to ?1 dB). This clearly shows that, with the 2A3’s 800 ? internal resistance, a transformer with a theoretical primary inductance of 20 H is enough to achieve a response down to 12 Hz at ?1 dB. Now, let’s see what happens if we replace the 800 ? with the 22.5 k? of the 6L6. In practice, we’re simulating pairing the 6L6 with this transformer designed for the 2A3.

In this case, ?1 dB occurs at 363 Hz. We’ve basically chopped off the bass. It might seem like a good way to avoid overly intrusive lows in a zero-feedback amplifier, albeit probably not this drastically. However, it’s important to know that this practice is fairly common among more experienced zero-feedback amp builders. The best among them adopt transformers with high primary inductances but designed at the threshold of core saturation. In this way, below a certain frequency the transformer fails to amplify correctly—sacrificing frequency response in favor of distortion that can be quite evident. The waveform in these cases tends to become almost unrecognizable—and indeed, that’s exactly what happens.

Nevertheless, this technique is also used by some very well-known names in audio, although I prefer not to cite them to avoid prompting unnecessary commentary from some “experts.” Returning to our theoretical transformer, how can we make it work with the 6L6? A solution that might seem obvious would be to increase the primary inductance. Let’s see what happens.

Here are our 12 Hz at ?1 dB with a primary inductance of “only” 550 H. It doesn’t take much to build a transformer with 550 H of primary inductance, right? Naturally, there will be the usual “gurus” with brilliant smiles and gold teeth, talking up exotic core materials and selling transformers with such inductance figures on paper. Too bad that, when measured, they often come in well below 100 H.

Moreover, some will tout toroidals, round cores, or other closed low-loss core configurations originally designed for power transformers. These can indeed show very high inductances when measured in the lab with no DC in the core. In practice, however, these transformers begin to show distorted sine waves below 100 Hz (I’ve seen it with my own eyes) due to core saturation, especially in single-ended outputs. Even in push-pull, if there’s any slight bias imbalance, the problem appears.

The usual makers, in their view, claim these transformers produce “more articulated bass,” but the reality is they generate distorted bass. But that’s fine, right? It’s a zero-feedback amplifier: according to their philosophy, the bass must be “choked,” otherwise it becomes too intrusive.

The reality is that to get a high primary inductance you either need a top-quality core or a large number of turns of copper. High-quality cores do exist, but as I’ve said, they don’t perform miracles. Another factor to consider is the repeatability of the transformer. Even if the material doesn’t come from scrap iron but from good-quality stock, there are still issues to consider.

The reality is that to obtain high primary inductance you must either use superior cores or increase the number of turns. High-quality cores exist, but they don’t work miracles: using them only makes sense in specific applications. It is also important to consider the difficulty of maintaining homogeneity between core batches, even within the same production.

In audio—being a niche—transformer manufacturers do not invest significant resources in R&D, since the market is much larger for industrial applications than for audio transformers. This leads to products which, despite enticing technical claims (e.g., sky-high theoretical inductances and exceptional frequency responses), do not actually deliver the expected performance. In short, “super” cores should be used only when justified, without blindly trusting technical statements that often do not reflect real-world results.

Figure: Frequency-response graph of a single-ended interstage transformer for 6SN7/6J5. Although the very famous manufacturer claimed 300 H of primary inductance and a 5 Hz–40 kHz ±0.5 dB response, tests reveal far inferior performance, with only 40 H and a ?1 dB attenuation already occurring at 7 kHz. The manufacturer, when confronted with measurements proving the product didn’t work, refused a return.

Returning to our 2A3 and 6L6 analyses, it’s evident that a transformer designed for the 2A3 might not be suitable for the 6L6. In fact, the 6L6 requires a much higher primary inductance than the 2A3. We have already highlighted how building transformers with very high primary inductance using ultra-high-permeability cores is feasible, but only up to a point—beyond which further problems arise. Another approach to increase inductance is to use a larger core and, as already mentioned, to increase the number of turns. These strategies offer alternative ways to meet the 6L6’s requirements compared to those optimized for the 2A3.

However, increasing the number of turns has a drawback: the winding’s DC resistance increases, and part of the available power is inevitably dissipated across it. Using a larger core also entails losses: part of the energy is dissipated or dispersed within the magnetic material itself. For these reasons, although on paper a 6L6 should deliver 6.5 watts, in practice—once inserted into a real circuit—it may deliver only 5 watts. The same holds, for example, for an EL34, which according to the datasheet should yield 11 watts in single-ended, but in practice hardly exceeds 7 watts actual. so much for those who sell gear with a single EL34 that they claim delivers 15 watts.

I’d also like to highlight an interesting aspect of many vintage devices, such as radios or classic amplifiers: you often notice that their output power seems higher than what you get from modern, even well-designed projects or high-fidelity builds. Looking more closely, you see that the output transformers used in those devices were much more compact. Less copper means lower DC resistance; less iron means lower magnetic losses; hence a larger share of the generated power actually reaches the speaker.

The flip side is that these transformers offered limited frequency response: typically 500 Hz–5 kHz at ?3 dB, sometimes even less. Some more “advanced” models, like certain Harman Kardon, Scott, and the like, could reach 300 Hz–8 kHz at ?3 dB, but with significant phase rotation—real roller coasters. In short, they had a sonic character of their own, which some might even like, but it’s best to be aware of their objective technical limits, especially when compared with modern hi-fi amplifiers designed for a much more linear response.

There’s another important aspect to consider. With the SPICE simulation example we analyzed the behavior of an ideal inductance in series with a resistor, but a real transformer is a far more complex component. Beyond the primary inductance and its DC resistance, a number of other non-negligible elements come into play:

• the internal parasitic capacitances of the primary and secondary
• the capacitance between primary and secondary
• the overall capacitance between windings and core
• the possibility that the core is mechanically—and thus electrically—bonded to ground (for example if bolted to the chassis)
• not to mention the system’s own resonances, leakage inductance, and other complications often overlooked in simplified models

In short, the transformer is far from an “ideal” component: it is intrinsically complex and full of pitfalls that must be understood, evaluated, and managed carefully, especially when you aim for sound quality and circuit stability.

As we have seen so far, the low-frequency response is closely tied to the ratio between the tube’s internal resistance and the transformer’s primary inductance. In the presence of pentodes with high Ri, it becomes clear that, regardless of all possible solutions, the use of negative feedback often remains the only effective way to get decent bass response, because it reduces the effective Ri seen by the transformer.

In our SPICE example, it also appeared that the high-frequency response was theoretically unlimited. But it must be emphasized: that was an ideal circuit. In reality, the upper passband is heavily influenced by parasitic capacitances inside the transformer. It’s tempting to think that a triode, thanks to its low Ri, is always favored over a pentode, but experience shows it’s not that simple.

With the same transformer, I have often observed that tubes with higher Ri do lose something in the lows, but can extend higher in frequency than those with lower Ri. Moreover, the more you push primary inductance up in an attempt to improve bass response, the more you sacrifice the top end, to the point where even high-Ri tubes start to suffer.

In practice, this complexity boils down to a key rule: it’s not always advantageous to increase primary inductance too much, because it can become counterproductive.

This is why, even if on paper a 2A3 and a 6L6 may require the same nominal 2.5k load, the transformers intended for one or the other will never truly be interchangeable. Even if the core could be the same, the number of turns—and thus the entire winding design—will still be different, tailored to the specific characteristics of the tube.

In conclusion, a well-designed transformer should be optimized at least for the specific tube you intend to use. It’s true that you can sometimes use common transformers for similar tubes—for example, a SE around 5k can suit EL84, 6V6, or PCL82—but this works only within certain limits: similar currents, power between 2 and 5 watts, and compatible tube characteristics.

Otherwise, more dissimilar tubes require specific adaptations. A 6600 push-pull can work for both EL34 and 6L6, but in single-ended an EL34 works better on 2k, while a 6L6 prefers 4k2. This is precisely where many “universal” low-cost transformers fall short, marketed as compatible with everything—from 6V6 to KT88, up to 2A3 and 300B—but in reality optimal for none of them, causing poor performance even with feedback applied.

And no, a 2A3 does not need an ultralinear tap, because it’s a triode.

My point is simple: building a tube amp with cheap components is never truly economical. If I’m going to build, I demand maximum quality; otherwise, I might as well use a 20€ Chinese class-D amp.

I cannot and will not compete on price, because I am not a factory, but I outsource transformer production to external companies that follow my strict specifications. I also offer services like a custom schematic (+€50), which requires hours of work but provides a tailored, optimized design.

My approach to quality: how all audio transformers should be

There are excellent industrial winders out there, capable of making three-phase, single-phase, very high-voltage, medical, and military transformers: all power transformers, working at 50 Hz or in switching. On that front, nothing to say: competent, reliable, and professional.

The problem arises when moving from power to audio. For a power transformer, it’s enough to check that the voltages are correct, it doesn’t get too hot, and insulation holds: if all that passes, the job is done. But an audio transformer is a different story. Everyone “knows how to calculate it,” at least in theory—you’ll find it in any book on the topic—but the leap from theory to practice is huge.

The real issue is the transition from theory to practice. When I build a transformer, there are many variables to weigh: I can increase current density, change core size, or induction frequency to obtain the same theoretical result. But in the end, quality shows only when the transformer is field-tested.

Some charlatans claim that “an output transformer is a static machine,” as if a transformer’s quality didn’t depend on its maker. In reality, there are many ways to build a transformer, and the final quality depends on the competence of the person who makes it. Not all transformers are equal and cheaper ones don’t always offer optimal performance. If you’re looking for a high-quality output transformer for tubes, you won’t find it in every product—especially not in the cheapest ones.

How do you test an audio transformer? Unlike a power transformer, which you can test easily with a load resistor, an audio transformer requires a more precise approach. Some professional winders use a function generator and an oscilloscope to observe frequency response, but without a load, this test is partial. They show the sine wave dropping “a bit” at 20 kHz with 20 mV out, but without a real load you don’t get an accurate idea of the transformer’s behavior under operating conditions.

Expensive instruments for measuring passband never fully simulate real-world operating conditions. Therefore, while they can be useful during development, the real test only happens in the actual circuit with the tube.

In my lab, I wind and test transformer prototypes myself, making corrections based on the results. One example is the transformer I’m designing for headphones and as an interstage with a 101D tube, with a 12k primary and 250-ohm secondary, intended to handle low power (up to 250 mW).

The first specimen, with the black bobbin, was a partial failure. When connected to its actual tube, the frequency response showed ?1 dB just under 20 kHz. Many would consider that a good result, but for me it was unacceptable and had to be scrapped.

Even though the first bobbin ended up in e-waste, it still gave me useful data. I could tweak some parameters without straying too far from the initial design. So, I wound a second specimen, slightly reducing the primary turns and the primary inductance. This unit, as you see below, was then connected to the only measuring instrument I consider reliable: the tube I want the transformer to work with.

Here is the frequency response of the second prototype, showing an excellent ?1 dB a little over 25 kHz (circuit without feedback).

In the photo, the potted 101D interstage, which for the curious has only 65 H of primary inductance… and that’s perfectly fine.

Here’s another example:

In these photos we see prototyping of a transformer for 32-ohm headphones, designed to simplify the process. I used a pentode dissipating almost 8 watts in triode configuration (a 6CL6, though a 6H6Pi or 6H30 would also have worked), paired with a transformer with a 2500-ohm primary impedance. This transformer, with a “humble” core and about 20 H of primary inductance, can deliver over 100 mW into 32 ohms, with an acceptable passband of about 25 Hz–30 kHz at ?1 dB.

And so I continue to hand-prototype and test using the actual tube—which I consider the best instrument. Once I finalize the winding design, I hand it off to the external firm that handles series production, applying all necessary checks and labeling.

My instruments

The problem today is the decline in the quality of online discussions, where many people I call “trolls” or “keyboard warriors” seem to regress instead of grow. In the past, I measured transformer frequency response with an oscilloscope, publishing only verifiable data, yet sometimes I was accused of falsifying results. To improve, I acquired a USB instrument that lets me plot and save frequency-response graphs, but the criticism continued, claiming I edited the graphs in Photoshop.

Even when I make accurate measurements and publish clear data, some keep challenging everything, accusing me of manipulating results. I have always clearly stated the test specs, but such baseless criticism is hard to avoid. It’s frustrating when someone criticizes my products while other companies sell questionable items at exorbitant prices without any reproach.

Many prefer to lower the bar instead of trying to improve. It’s a problem—especially in Italy—where mediocrity often prevails. Today, on social media and in electronics hobby groups, there’s often no room for constructive debate. Criticism is seen as a personal attack, and those who don’t conform to groupthink are marginalized. In a world where quality is often ignored.