Antipodes Audio K50 music server

On the face of it, playing and streaming digital music files is a straightforward process. You direct data from various sources—some local, some "in the cloud"—perhaps via a reclocker/signal conditioner to a digital-to-analog converter (DAC). "And the music comes out here."

Not so simple. Bits, it seems, aren't bits, or not only. A digital datastream is also an analog signal. Noise and other signal errors endemic to multi-function computers not designed primarily for music playback can affect how music sounds. And then there are the practical issues of setting up and connecting everything optimally, and then organizing music files correctly, which can be especially difficult when ripping files from multidisc sets.

These obstacles are why so many audiophiles have either switched to one-box music server solutions or even thrown up their hands and stuck with physical media.

Ever since I began using my Roon Nucleus+ music server, powered by an HDPlex 300 linear power supply rather than Roon's supplied switching power supply wall wart, I've wondered about the quality of other server/streamers that cost considerably more. I've listened to a few, including the Wolf Audio Systems Alpha 3 SX and the Innuos Statement.

From the Antipodes
In June, I requested a review sample of my most expensive yet, the one-box, flagship server made by New Zealand company Antipodes Audio, the K50 ($15,000, footnote 1). My contacts at the company were Mark Jenkins, Antipodes's CEO, and Mark Cole, the company's head of service, sales and marketing. Which meant that, in addition to figuring out how to operate a unique device with multiple choices of inputs, outputs, servers, and players, I had to figure out which Mark was which.

Jenkins, who was raised on Bizet, Puccini, Gershwin, Rimsky-Korsakov, Borodin, and the Rolling Stones, spent several decades in the digital transmission, broadcasting, and telecommunications fields before he entered the hi-fi industry in 2004. He focused on cables at first and then turned to music servers.

"I was struck with a conundrum back in the '90s," Jenkins explained in an email; we also communicated extensively in WhatsApp. "Using computer audio opened up my library of music to me in a way that playing a record or a CD did not. Yet the record and CD sounded better. ... This was despite me using a couple of different reclockers and some high-end DACs that claimed to fix jitter completely [and] render the limitations of the source irrelevant. ... So I decided that the simple explanation that 'data is just data' needed to be examined."

Experimentation led Jenkins to the belief that every component in a stereo system matters. "For computer audio, the idea that you can do a bad job at the source and fix it later is simply out of line with what I hear. ... Many of our competitors focus on either a low-cost or a high-power computer to crunch numbers; then they rely on things like slow linear power supplies and add-on devices designed to filter noise or regenerate the signal to fix things after the event. ... [Antipodes] starts with audio design from the ground up.

"As explained on our technology page, we believe the industry has focused too heavily on noise and adopted noise reduction techniques that compromise bandwidth." Instead, Antipodes aims to ensure a high-quality digital signal from the outset and to preserve bandwidth. "We believe this sets the sound of our music servers well apart from those of our competitors.

"The problems of early digital have been leading people in the wrong direction. In the beginning, we were largely trying to get rid of the harshness and nasty top-end noise that came about through (a) using hard drives and (b) interference with the audio signal. A lot of the way that people dealt with that noise was to slow everything down with slow linear power supplies, noise filters, and regeneration, which tends to act like a filter as well. The fundamental difference between our approach and what we see most of our competitors doing is that we approach everything from the ground up rather than applying Band-Aids after the event. I think we are also unique in our focus on bandwidth as well as noise rather than focusing exclusively on noise.


"Our goal is to provide a perfect squarewave to the DAC, ... [one that] is able to turn around a corner at a perfect right angle and doesn't have wiggles and squiggles that confuse timing. ... We started to work on improving bandwidth rather than reducing noise [because] if we improved bandwidth, then life, verve, vigor and emotional involvement—the stuff that makes you smile, cry, or dance—came through. ... If we allowed noise at the beginning of the signal, we would have to start filtering and slowing things down to get rid of the problem." Instead of "starting with rubbish, we design from the ground up to minimize noise. Because of that, we can also maximize bandwidth."

In one of numerous follow-up chats with the two Marks, Jenkins noted that worldwide chip shortages had forced a few changes to the K50's design and that the beta software I'd be receiving wouldn't be bug-free (footnote 2). Nor was an updated manual available. What was consistent with the previous iteration of K50s, however, was that it honors Jenkins's "personal bias that timbre must be accurate" for music to have life.

"The ultimate goal of Antipodes Audio is to deliver the emotional fulfilment that can be hard to achieve in the course of an ordinary day," he said. "Every now and then, I listen to a distributor's $130,000 turntable setup and ask myself, 'What are we missing?' The great turntables have a kind of life and verve to them that the music servers of a few years ago just didn't get. Getting that life and verve into an Antipodes music server has been my biggest goal over the last few years. ... It's easy to get a music server to throw loads of detail at you; getting it to make musical sense is the hard part."

Antipodes optimizes the K50's power supplies to increase speed and lower noise. The company works with two manufacturers of industrial motherboards to ensure optimal tuning for precision and noise performance. "Tuning is really only available to a motherboard manufacturer," Jenkins said. "With their help, we can optimize audio performance. This gets us high-quality, great-sounding motherboards, albeit at a relatively high cost compared to using standard motherboards.

"Your reference Roon Nucleus+ is not quite the same thing as the K50. The Nucleus is there to play the Roon server app; the K50 plays a range of server apps on one internal device and the selected player app on a second, isolated internal device. It also generates a full range of digital outputs using oven-controlled clocks.

"The K50 has three computational parts, each fed by a dedicated, overspecified, superfast low-noise power supply:

• The server computer runs the server apps, manages storage, manages internet streams. ... In simple terms, it provides high-quality streaming of files over Ethernet to the player computer.

• The player computer runs the player apps, outputting via USB, and feeds the [other] digital outputs. In simple terms, it receives the files streamed from the server computer and turns them into digital audio signals.

• The digital outputs unit is run by an FPGA computing engine and does a few things including providing galvanic isolation from the other computers, reclocking the digital signals by means of an ultrahigh-quality, oven-controlled clock, and generating signals in appropriate formats for each digital output. It is worth noting that the conventional digital outputs (S/PDIF, AES3, I2S) need to be generated using a high-quality clock. This is less important for USB and much less important for Ethernet, where issues like noise and bandwidth assume greater importance."


The K50 can work with Roon as server and player, Roon as server with Squeezelite or HQ Player as player, or Squeeze as server and Squeezelite as player (footnote 3). Jenkins refused to advocate one or the other in print, although he did acknowledge preferences. After strongly encouraging me to try everything I could, he said, "If you're used to Roon, that's a good place to start. But [every server/player] combination sounds different, and preferences can change as the different software changes.

Footnote 1: JA reviewed the company's previous flagship, the DX Reference, in October 2015.

Footnote 2: This was an understatement.

Footnote 3: Squeeze and Squeezelite are the native apps for the Logitech Media Server, aka Squeezebox.

Antipodes Audio Limited
Suite 1, 2 Ake Ake Place
Otaki 5512
New Zealand
(303) 495-2260 x116

JGlacken's picture

Would you be good enough to compare the two above please

Jason Victor Serinus's picture

Given multiple changes to my digital set-up between "then" and "now," I'm most comfortable saying that what stood out for me most about the Innuos Statement "then" was its warm and smooth signature.

thatguy's picture

"Squeeze and Squeezelite are the native apps for the Logitech Media Server, aka Squeezebox."

The good old Squeezebox players; gone but not forgotten and still widely supported. I've been using them since the original slim devices days in 2005.

They were just a bit too ahead of their time. People didn't know they wanted high quality access to their complete music system with multiple control options yet. I got strange looks when I'd try to explain it to people in 2005.
I still prefer the squeezebox touch to most new options; it is nice having the choice of using a touchscreen, a remote, an app or a web browser to pick what music for listening. Add your favorite DAC and the sound is top notch.

georgehifi's picture

You still have no knowledge of the provenance of what's being streamed to you.
Is it the early uncompressed or later compressed released version of the same album.


And the Boss.

Cheers George

Jack L's picture


Bingo !

Being a vinyl addict, I would never want to spend $15,000 for a music server though I often stream to watch HD classical performances on my 4KUHD TV hooked up to my tube amps for their HD sound.

I am impressed to learn, for the first time ever, a hi-end music server maker top guy quoted putting "emotional fulfilment" into his new server "has been his biggest goal over the last few years".

Yes, he said he was inspired by the "emotional fulfilment" he sensed while auditioning a $130,000 turntable set up. I shared similar "emotional" sensation as I also auditioned the USD125,000 Audio Note Japan turntable system years back.

So this $15,000 music server should sound musical close to its pure analogue counterpart: turntable, hopefully !

Listening is believing

Jack L

Archimago's picture

Footnote 5: Jenkins wrote, "I should add that I am not a big fan of high-bitrate recordings or DSD. ... This is just a personal view, and nothing to do with any design choices we have made. Bigger bitrate numbers do not mean 'better.' High bitrates require longer filters, which makes the top-end smoother but shifts the problem down further into the audio spectrum. On top of this timing errors are magnified, transmission has to be accurate at even higher frequencies, and asking any computing device to operate at a higher speed means it operates at reduced precision. ..."

Don't know if "high bitrate" is referring to higher samplerate here (it must if we're talking about lossless).

Higher samplerate material will actually not require "longer filters". In fact, we can get away with more relaxed filters and even go filterless for example at 192kHz and 384kHz since artifacts will be quite far ultrasonic. If you use a NOS DAC, and feed it high samplerate audio data, you achieve better quality for example compared to 44.1/48kHz unfiltered.

While older interfaces like AES3 and S/PDIF might struggle at 24/192 depending on the quality of the streamer and DAC, certainly with high speed USB 2.0 and other asynchronous interfaces like ethernet, this will not be an issue.

I don't know why this is even a "personal view" since these things can be quite easily demonstrated.

gasman's picture

Archimago is absolutely correct (as always) which raises the alarming prospect that Mr Jenkins either doesn't know his subject in sufficient detail to maintain credibility or is spurting nonsense because he thinks the gullible audiophile won't know any better. Neither option is particularly reassuring from someone trying to sell a tricked out PC for $15k!!

CG's picture

Well, almost.

Even though those artifacts are ultrasonic, that does not mean that they can't or won't cause problems. Almost all analog amplifiers become far less linear at ultrasonic frequencies, so intermodulation products from those ultrasonic artifacts can and will mix down to the audible range. This can be shown mathematically as well as through testing.

High sample rate audio data fed into a NOS DAC is ideal, as long as you apply some analog filtering after the converter to get rid of the ultrasonic conversion products. Because of the high sample rate, that filter can start rolling off way above the audio band and be relatively low order and still be effective.

Other than that one nit, I really like your solution. Bet that makes your day...

ok's picture

..he actually means "longer (slower) roll-off filters" -in contrast to brickwalls for instance- since that way his overall statement makes perfect sense. Having said that I still don't get what a server/streamer is all about. A small portion of this kind of cash buys you a top-end laptop with every imaginable sound-oriented upgrade which does everything a server/streamer does and then some; plus, in case it leaves you wanting, you are still left with a top-end laptop and some serious cash.

Pork Needle's picture

<< Noise and other signal errors endemic to multi-function computers not designed primarily for music playback can affect how music sounds.<<

Nonsense. A massive Photoshop file duplicated a million times will be EXACTLY the same as the original. A music file is no different.

hb72's picture

.. it is interesting to hear how different one & the same file can sound, if e.g. only the power supply to a streamer bridge is improved (well, in my system it is). This doesn't influence or improve "the file" but somehow enables e.g. the DAC to do its job in a better way (simply speaking), possibly by being less influenced by noisy ground, jitter, or any other real-world nastiness.

I would not relate the word (digital) "signal error" to file corruption or "file error".

In my recent experience, the number of files that do not end up corrupted after copying is satisfyingly large, but definitely smaller than a million, before one corrupted copy is produced, though I really hope your stats are better than mine. :)

gasman's picture

Another interesting experiment along the same lines , whilst running music through Roon I give my server other CPU intensive jobs to do. Toggling these jobs on and off makes no discernible difference to the music and I can genuinely A/B this. So if taking CPU from 5% utilisation to >60% doesn't effect the sound, what role is the server (not the player) having?

CG's picture

That is certainly true, but really isn't the point.

The conversion back to analog isn't perfect. Even assuming that the bits are perfect (let's say that is so), the process is full of noise. There's logic gates toggling, power supplies powering, and so on. This can be measured and is true not just for audio systems but for any mixed signal processing system.

To make it all worse, the various boxes used are connected not only by the "interconnect" cables, whether digital or analog, but also through the AC mains. So, common mode noise finds its way pretty much everywhere. (For more information about this aspect, I'd suggest finding articles by Henry Ott and Ralph Morrison. Easy to search for.)

People can and do measure this stuff. For some reason, the audio world avoids it all in favor of arguing. Can't tell you why.

skikirkwood's picture

I would love to see a follow-up article comparing the sound of this $15K streamer to a Raspberry Pi running Volumio or piCorePlayer. Both support Squeezelite.

Blind A/B test of course.

Archimago's picture

As logical and useful as this would be, somehow I don't think this is going to happen in the pages of the current Stereophile. Maybe back in the day when there used to me more investigative articles to help hobbyists understand the technology.

A shame. Hopefully one day mainstream magazines could return to the roots of helping readers make sense of what is or is not important rather than taking all claims basically at face value with no apparent ability to show skeptical inquiry given obvious facts about how the tech works.

skikirkwood's picture

A/B a $100 solution that sounds exactly the same as a $15K alternative? The result would end the writer's career if published.

The most entertaining part of the article was this quote:
"It takes a K50 about 2 months to burn in properly," he told me. "It sounds too thin and sharp for the first 200 hours. If you turn it off for any length of time, it will take about 3 days to get back to optimum performance."

Kal Rubinson's picture

Perhaps they should offer it with a battery back-up..................

skikirkwood's picture

Good one! Of course Raspberry Pi's can run on batteries and that's a lot cheaper than high-end linear power supplies.

hb72's picture

I assume one can hear the difference already from another room.

clock timing is an issue, power supply quality, separation of good old usb vs network is an issue, so that I'd suggest to include also Volumio Primo* and Stack Audio Bridge or, as an alternative to the latter, the almost identical pro-ject stream s2 ultra*, to show what technical improvements in mentioned points would offer in terms of SQ.

*) both tested here on Stereophile, thanks!

Mikernz's picture

Always such interesting music, JVS. Where can I get a 24/96 remastering of this great classic?

Jason Victor Serinus's picture

I have the files of the 2011 remaster in 24/96, but the three download sources I've checked so far only offer 16/44.1. They must have withdrawn the 24/96 download, for reasons unknown. You can stream it in 16/44.1 MQA from Tidal, however; IMHO, that is the best you can do sonically right now.

Maybe Warner has plans... ?????


navr's picture

How is bandwidth increased within the Antipodes server?
Are there still Antipodes cables with cotton based outer layer, allegedly the only cables that do not have plastic within the EM field around the wire (which is made of alloy of gold, closer to the surface, and silver, farther from the surface)? Or these cables lost the battle to other brands (that still have plastic within the EM field even though air comprise the most of the gap between the plastic layer and wire)?
Since audiophile switches must have some filtering within thus slowness, what is the Antipodes stance toward these switches (which were measured by ASR using Audio Precision instrumentation and found no difference between them and $20 switch, though the measurement was only in the frequency domain i.e. integration operation). The philosophy of Antipodes and (independent) measurements by ASR seems to complement each other.
In the light of above, which particular Ethernet, USB, and other digital cables are recommended by Antipodes between router and switch, switch and local server, local server and player, player and re-clocker, re-clocker and DAC?
In line with the overall philosophy, are the DACs with very wide bandwidth and re-clocking duties delegated to the upstream re-clocker preferable over expensive DACs that assume player and re-clocker duties on top of DAC duties?