2023, Chapter 9:
Back in Black
Mjolnir has always been an interesting amp.
First conceived as an affordable, balanced, stop-gap amp before Ragnarok’s introduction, and then totally switched up as a balanced tube hybrid that could also go solid-state, Mjolnir is now back as a third generation—and, true to form, totally different again.
Huh? Are we nuts?
Don’t we know that different generations of the same product usually just tweak a few details, and call it a day?
Sure. That’s fine. But for an amp that takes up a significant part of your desktop, maybe it’s best as an experiment, as something that pushes the limits, as…ah, to be frank, not a regular production option.
So welcome the new limited-run Mjolnir 3.
By far the craziest Mjolnir yet.
All Out
“You gotta be kidding,” someone says. “The craziest Mjolnir? The first was a circlotron and didn’t have single-ended outputs. The second had tubes and introduced LISST. How much crazier can you get?”
This crazy: real Class A.
Aside: and for what me mean by “real class A,” I wrote a chapter on it here.
Oh yeah, and choke-input. And true dual mono (yes, 2 transformers, 2 chokes). And 100% discrete, including power supply. And it uses 32 10A-rated output devices (no, not a typo—32 10 amp rated output devices)…for a headphone amp!
And runs so hot it needs 2 Gjallarhorn heatsinks.
Aside: Gjallarhorn is our 2 x 10W speaker amp. Mjolnir 3 needs 2 of its heatsinks. Yes. Two.
Oh yeah, and switchable Class A mode—single-ended or push-pull. And switchable feedback—none or yeah. Oh, and switchable gain that’s independent of feedback.
“Wait wait wait,” you say. “Switchable mode? Switchable feedback? What does that mean?”
It means that you can pick your Class A mode:
- Class A single-ended
- Class A push-pull
You can also choose your feedback:
So yeah, you can totally have a Class A single-ended no-feedback power amp…or change it up if you need more power or lower output impedance or whatever.
You also get switchable gain, switchable preamp outputs, balanced and SE inputs, balanced and SE outputs, all hot-switchable, all solid-state, no tubes, no transformer outs, all DC-in and DC-out. Think of it as a solid state Folkvangr—that is less expensive and more flexible.
In short, there’s never been an amp like this.
R&D…&D…&D…&D
“So how did you come up with this crazy idea?” you ask.
In short: we didn’t.
Or at least not how you’d expect.
At first, I thought Mjolnir 3 would be something much more, ah, normal. As in, hey, we have this Nexus™ topology now, and it is better than a circlotron in so many ways, so why not do a Mjolnir 3 with tubes in Nexus? I mean, we hadn’t done tubes in Nexus.
So that totally made sense.
Except tubes in Nexus totally didn’t work.
They didn’t work so badly that we still don’t have a tube Nexus product. We’ve taken a number of shots at it, and the problem always comes down to the input devices needing to be tightly matched.
And tubes, even on the best of days, are not tightly matched.
And, even if we tightly match them, over time they will drift apart.
And if we cheat, say, like doing a cascode input with tubes on the top, is that really tube gain?
Ah, the pain we cause ourselves by being principled! It would be soooooo much easier if we could just fake it. Or just throw a bunch of opamps in a box. Oh wait, we did that. But it’s cheap so maybe that’s OK. I’ll fix that here soon anyway.
But back to Mjolnir prototypes.
When I first got Mjolnir Nexus Tube 1 running, it only ran for about a minute. Then it went out of Nexus and went out of balance, with high distortion.
Well, that sucked, I thought. I swapped the tubes to another pair. This one never even ran. It just heeled over to one side and lay there, unable to balance in Nexus.
Back to the first tubes. It ran for a minute, fell over on its side, then, over time, slowly kinda came back to a sorta-ish balance, then fell over again.
Aside: all this “balance” and “falling over” refer to the fact that it takes a pair of operational point servos to get Nexus to work at all, and it works by balancing on a knife-edge. This is why we have to match input JFETs to 1% (not a typo—not 10%, but 1%), match all gain elements, and observe close thermal coupling so everything stays in the center operating point. You’re never going to match tubes to 1%--and, even if you did, it’s doubtful they’d stay that way…maybe not even for a month.
“Well, that really really sucks, “ I said, out loud, and put away the proto for a while.
Eventually, I brought it back out again and tried it with LISST. This worked. So I went back to tubes. Nope. I matched tubes as close as I could. I got a kinda-sorta-leaning-to-one-side Nexus. Not confidence inspiring. I tried pasting in a differential balancing servo, but the many time constants involved in two operation point servos and a balance servo meant it would still sometimes drift out of alignment…and sometimes it might take days.
And that was just the beginning. There were a lot more protos to come.
So how many Mjolnir 3 prototypes did we make? A lot.
- Mjolnir 3 Tube Nexus 1. The first, falling-on-its-side kind of proto, as mentioned above.
- Mjolnir 3 Tube Nexus 2. An improved version with better thermal, higher rails, cascoded 2nd stage, integrated and improved balance servo. Nope. Same thing. Fell over on its side after 10 minutes to 3 days. Not reliable enough to produce.
- Mjolnir 3 Nexus Tech Leader. After that, I figured “hey, if tubes are the problem, we can treat this as a tech leader thing, combine Nexus and the Programmable Output Stage (a super-Continuity that eventually showed up in Magni Piety) with relay ladders and a microprocessor and go from there. This one was okay, but it was: (a) too close to Ragnarok 2 in basic idea, and (b) Supplanted by Jotunheim 2
- Mjolnir 3 Old Skool Circlotron. Yeah, we did just and improved Mjolnir 2. It was okay, but it was, well, just another Mjolnir 2. And yeah, I know, Mjolnir 2 is very nice. I have one myself. But it was time for something different.
- Mjolnir 3 Crazy Simple Real Class A. And if we wanted to go different, I figured, maybe real different was the way to go. As in, a full crazy-stupid, real-deal, old-skool Class A (with chokes, because hell yeah), single-ended, heck, just do one good rail and cap-couple the output, and concentrate on making the single rail the best it could be. Yep. It worked. Fairly well. A bit noisy. And a bit limited with that single rail. No, what we needed was something with a big-ass rail for voltage gain, and lots of current for output. Kinda like the first Mjolnir. And if we did split rails for output, we could get rid of the output caps. And we could see if we could switch it to push-pull. Which led to…
- Mjolnir 3 Still Simple Still Real Class A. Where we are today. This takes the best of #5, combines it with some ideas from the original Mjolnir (high voltage gain stage), refines it, and packages it into a big, hot, heavy hammer of an amp. We’ll talk more about this is a bit.
But all of the above really undersells the level of crazy in R&D on this one. In the midst of this, we were considering a smaller form factor, so some of the prototypes above were designed for a 12” chassis, some were designed for a 14.5” chassis, some took cards like Jotunheim and Ragnarok, some had other tricks we learned a few things from and may use someday…but in the end, we went back to the standard chassis size and no cards, to keep it simple (and to forestall any complaints about stackability, and to ensure we didn’t need to do a whole new class of shipping package).
It also really undersells how bad most of these prototypes were. None of the iterations, up to #6, was worth listening to for any length of time—either due to noise, unreliability, operational issues, oscillation, or just plain crappy sound. The tube Nexus version, when it worked, was actually more stable and measured better than the tech leader. This didn’t mean it was amazing—it meant there was something wrong with the tech leader. I tried to chase it down…but never really got it 100%.
And the whole card thing—yeah, what a can of worms. One concept had two card slots. One had a new big slot for new big crazy cards. In both cases, we considered it over the top.
But even when we ended up at #6, it didn’t mean that development was over.
Oh no. Not by a long shot.
The Dev is in the Details
There were three separate versions of #6 built. The first one had significant challenges (engineering code for,
wow, I sure screwed up this board). First and foremost was that the voltages were reversed to every single relay. I really wanted to listen to this stupid thing, so I dremeled out all of the traces to every relay and ran wires in the air to correct it.
Aaannnnnd…it worked. Pretty much. A couple of the things I was worried about turned out to be no big deal. The biasing worked just fine. The push-pull switch worked just fine. I’d even guessed the right resistor value for the push-pull switchover (super fancy thin-film resistors, ones that cost 300x what a normal resistor costs, hey yeah there are some fancy-ass parts in Mjolnir 3).
But we had other problems, most specifically noise. The amp was wayyy noisier than expected. I mean, yeah, a single-ended Class A design with no feedback isn’t going to win awards, but it should be quiet, especially with chokes in the power supply.
Turns out the chokes were the first problem. I’d under-specced them at 50mH. I really needed more like 200mH. So I had those re-wound with the directive of “max inductance at specified resistance, never mind the core,” which resulted in two surprising outcomes:
- The chokes shrank. They actually performed better on 24VA cores than 48VA cores. Weird, but OK.
- The inductance went wayyyy up. As in, the production chokes are more like 400mH. So 8x the original chokes.
For those of you wondering why this matters: more inductance means more energy storage which means more DC smoothing which means lower noise which means more audiophile woo to smear on the product.
Aside: I know, I’m being cheeky in saying “audiophile woo,” but I really can’t tell you why chokes are measurably better than capacitors for filtering in a power supply, so it has to come down to something that’s either BS or something we don’t understand. I prefer the latter definition, some people prefer the former.
The good news: the chokes improved noise performance significantly. Mjolnir 3 still won’t win any awards for noise floor, but it’ll be usable on pretty much every full-sized headphone. If you’re using IEMs, ah come on, just don’t. Using them with a 15-lb amp that puts out so much heat it should have a grill attachment…are you kidding?
Buuuuuuut…Why?
Some people out there are still scratching their heads. “Why go through all this trouble? I mean, you admit this amp is big, and hot, and inconvenient, and not the best measuring, and not the most powerful…I don’t get it.”
Fair.
I mean, we live in an age where all it takes is a handful of TI’s latest op-amps, and you have state-of-the-art measured analog performance. At least for most things. If you want to run speakers, or if you want to do electrostatic headphones, you might need more. But if you want to top the charts of measured performance, it’s dead easy.
And, even if you don’t want to use op-amps, it’s dead easy to do a discrete design that has state-of-the-art performance. Discrete op-amp designs get you there fast. And, if you’re good, and persistent, and know a few tricks, then low loop gain discrete designs can give near-SOTA performance as well.
The thing is, those kind of designs pretty much all fall into a typical format: the Lin/Blameless Class-AB template, with overall feedback. A few venture into current feedback (like our stuff, starting with Magni+), but the vast, vast majority are voltage feedback, speaker-amp-style topologies.
And there’s nothing wrong with that. The reason that Lin/Blameless designs are popular is that they work well, and their pluses and minuses are well-understood.
But…there are sooooooooo many of them.
And they really were designed in the crucible of loudspeaker amps. They’re designed for low output impedance and high power. They’re designed for the needs of speakers.
But Mjolnir 3 isn’t.
Mjolnir 3 is designed for headphones, knowing there is a ton more freedom in designing headphone amps. We don’t need to worry about so much power. We don’t need huge damping factors. We can go over-the-top and do real Class A. We can eschew feedback. We can experiment with weird or simple topologies.
It’s the freedom to try all these wacky things that attracted me to headphone amps in the first place, now over 13 years ago. It was profoundly freeing to do stuff I never considered back at Sumo, when I was doing speaker amps. It’s no coincidence that our first headphone amps were a real Class-A no-feedback design, a all-triode no-feedback design, a high-power hybrid with no overall feedback and a weird take on a Taylor output stage, and a circlotron.
Mjolnir 3 carries on that tradition, but now with more than a decade of experience designing headphone amps, and learning a few things along the way. It’s a real no-kidding Class A SE no feedback amp—that you can also tweak to push-pull and feedback if you want to experiment. It’s over-the-top overkill for headphones, it has huge choke-input supplies and nary an IC in sight.
And I think experimentation like this is important. I think we need frontiers where we say, “Well, what if?” and discover new things.
And that’s why I did Mjolnir 3.
Ready to Get Hammered?
So is Mjolnir 3 for you? Probably not, to be honest. It’s a big, hot, heavy, desktop hog of an amplifier. It doesn’t have the best measurements or highest power output. It’s an unabashed salute to Class A, real Class A, and over-the-top in terms of being dual mono and choke input.
But, if you’re inclined… Mjolnir 3 provides an interesting option for all-out headphone amplification. It’s really not like anything else out there.
- It’s the only choke-input headphone amp we know about
- It doesn’t use tubes, so you don’t have to worry about tube life, tube performance, etc.
- It is really, no fooling, 100% class A
- It also allows you to experiment with gain, feedback, push-pull vs SE Class A, a etc.
- It’s not a shrunken Class AB speaker amp
- It’s really, no fooling, 100% dual mono
- It’s not a high-feedback, discrete op-amp design
- It’s not a bunch of op-amps stuck in a box
- It’s unapologetically mechanical and simple
If that sounds good to you, then maybe, just maybe, Mjolnir 3 is something you’d be interested in. If not, no worries. Most sane folks will go for Jotunheim.
But if you want to explore something new, something weird, something really off the beaten path…maybe it’s time for Mjolnir 3.
I hope you enjoy!
) with all the Schitty news.
Here Comes the Sun and Yellow Submarine are their favorites so far 
