I will offer my own perspective based on personal experiences that I hope will be helpful to those who are searching for their own music server. With so many server options available and with reviewers suggesting this server or that server is best, how do you choose?
I have been building my own music servers for years which isn't really saying much because almost anyone can buy an inexpensive NUC, install memory and an SSD, install the OS and a software player like Roon and voilà, you have a music server. But if this is as far as you decide to go with your music server, I won't be surprised if you find yourself reaching for your turntable or tape deck every time you want to seriously listen to music.
While at the Constellation room at the start of RMAF in 2017, Peter McGrath of Wilson Audio came by because the new Wilson Alexia 2s were in this room and he wanted to make sure everything was sounding right. The Constellation people queued up a digital track from their music server and Peter stopped them shortly after the track began playing and said something to the effect of "Let's get serious here and switch to the turntable."
This was not an isolated event. In fact, far from it. For a long time now, digital has been high-end audio's necessary stepchild, there for convenience but not really loved or respected.
When I would ask people what they felt was wrong with digital, the typical response would be that digital sounds harsh. Terms like "digital hash" or "digititus" would be thrown around as if the only problem with digital was noise. And so it seemed that the basic strategy of all server manufacturers was to reduce noise. After all, digital was just about a bunch of 1s and 0s and so nothing else mattered, right? This meant low-noise PSUs that sometimes consisted of high-impedance batteries and almost without exception, low-power CPUs because it was the general belief that low power CPUs = low noise. This meant CPUs like ARM, weak FPGAs, Celeron/Pentiums, or i3/i5/i7s that were sometimes de-throttled to achieve TDPs of well below 35w. If you don't believe me, check out the CPU being used by just about any server or streamer manufacturer out there and with very few exceptions, they will be using a low power CPU. Anytime you saw a CPU with a TDP rating of more than 35 watts, generally those servers were designed for HQPlayer upsampling and this includes Taiko Audio's own SGM 2015 and EVO as well as Pink Faun's 2.16X.
I fully bought into this strategy with my own builds. I devised the best power supplies I could find from the finest power supply builders I could find. I then married these power supplies to boards coupled with Celerons or Pentiums. Some of my servers consumed barely 10 watts and there was no question that these servers could sound clean and free of harshness but somehow they lacked soul.
By accident, I began experimenting with higher power CPUs with more cores and greater CPU frequency even though I had no interest in upsampling with HQPlayer and was quite surprised by the results even with these CPUs powered by inexpensive ATX power supplies. Yes, harshness increased but so did dynamics and staging. As I improved the quality of my power supply to the CPU and if I kept CPU temps to reasonable levels, the harshness went away and I was left with a signature that was not only more dynamic and expansive but fuller in body and with more realistic weight. In comparison, my Celeron server was now sounding thin and anemic. Perhaps even more importantly, music had greater immediacy and presence...and soul. This experiment has been repeated several times now and each time with the same results. Roon can sound so much better with a high-power CPU even during bit-perfect playback. As manufacturers start to do their own testing, I'm confident there is only one conclusion they can draw but this is easier said than done because it is much more difficult to design a good power supply for a high-power CPU rig. While the Extreme consumes only about 60 watts during music playback, Emile tells me that the Extreme can consume up to 300 watts during bootup. This would fry any of the bespoke PSUs I have on hand.
My experience with music servers are not only with DIY. I have owned or auditioned many commercially made servers or streamers from manufacturers like Sonore, SOtM, Auralic, CAD, TotalDAC, dCS, Antipodes, Aurender, and InnuOS. Some of these servers, especially InnuOS's former flagship, the Zenith SE Mk II which I used to own, were mated to an exceptional power supply but without exception, they all utilized low power CPUs.
During my time with the Zentih SE Mk II, I got to know some of the people within InnuOS and what I will say is they are the kind of people you would be happy to call your friends. During numerous conversations, I also found them to be technically very competent. As I opened up my Zenith SE Mk II, I found the build quality to be very high. At that time, it was among the finest servers I had owned and as good or better than any server I had built. It’s why I bought it.
When the Statement was announced, I was offered the opportunity to upgrade and I seriously considered it. Based on the provided specs, it seemed to tick all the boxes -- an even better power supply, a custom-designed motherboard, and high-level clocking to both the Ethernet and USB cards. But it was around this time that I had discovered the very positive impact of high-power CPUs and as the Statement was designed with a very low power Pentium processor, I knew it would be limited. Having heard the Statement on several occasions, while it is better than the Zenith SE that it replaces, it still bears the same thin and anemic signature that I now hear every time I hear a server powered by a low power CPU. While videos are not the best way to appreciate how well an audio component performs, I do find them useful when assessing for relative differences between 2 components and if you listen to the YouTube video recorded at Volent Audio in Hong Kong where the Statement is compared against the Extreme, while the recording is not of high quality, you’ll still get a sense of what I’m talking about, especially with the orchestral track. The Statement sounds clean but the signature is thin and the transients do not sound convincingly expressed.
If this is seen as a slight to InnuOS, it is not meant to be as this is an organization I am fond of. If you were to buy an InnuOS server, based on my own very positive experience, I predict you will receive friendly, attentive, and competent customer service, often from Nuno himself. Your server’s build quality will be at a very high standard and you will find only the best parts inside. The problem that ails the Statement is the same problem that most other m;usic servers and streamers being manufactured today have — the weak CPU. This applies to all the SOtM and Sonore endpoints. It applies to the CAD CAT I used to own. Same thing with the dCS Network Bridge. Same thing with Aurender’s latest W20SE which I had in my home recently. While I haven’t heard the Playback Designs MPT-8 that was mentioned recently, it appears the integrated server is based on a NUC. Regarding the Wadax Atlantis Server, if rumor is correct, that this server is also housing a NUC, then I see no possible way this can compete with the Extreme. Even if its power supply is as good as the Extreme’s and its RAM, networking, vibration control, and OS optimizations are equivalent to the Extreme’s, a low power NUC CPU is going to hold it back. In the case of a Roon server, low noise is important but performance is just as important and sometimes, performance requires horsepower. Throw racing tires onto a Toyota Prius, fill it with high octane racing fuel and place Tom Kristensen behind the wheel and you still aren’t going to win LeMans.
At this stage, the only legitimate competition I see the Extreme having is the Pink Faun 2.16X and the 8-core AMD CPU that it uses. While this machine was initially designed for HQP upsampling, when I spoke with Jord in Munich this past May, he said he personally doesn’t use the 2.16X for upsampling and that he feeds his DAC a bit-perfect signal using Roon because it sounds better that way. Like Emile, Jord believes wholeheartedly that bit-perfect Roon benefits from a high power CPU. In order to properly compete against the Extreme and it’s dual 10-core Xeons, however, you must combine dual 2.16Xs together in a server/endpoint configuration and indeed, such a configuration exists and is being marketed by Pink Faun as their best effort. In fact, I heard this combination performing beautifully in the Aries Cerat room in Munich this past May. The problem with this type of setup, however, is that even if SQ is equivalent (and that is a big “if”), dual 2.16Xs cost well north of $30k, consume more than 200 watts during playback, and very capably function as a room heater. This simply would not work for me.
At this point in time, I see the Extreme really having no peer, at least on paper, and the technology that has gone into the Extreme is what drew me to it. My trip to visit Emile was no coincidence as I needed to hear for myself if the execution was there and nothing I heard did anything to alter my opinion. It was as good as I was hoping for meaning it was the very best that I had yet heard, DIY or otherwise.