Music in the Round #96: Roon & Dolby Atmos

One of the recurring themes of this column has been my search for servers that will support the playback of high-resolution multichannel files with DSP for speaker/room equalization (EQ), as well as the format conversion and downsampling that are often part of those processes. Because most EQ software is PCM-based, format comversion is needed to convert DSD files to PCM. In addition, because most EQ products work within a limited range of sampling rates, PCM files sampled at high rates may have to be downsampled before being subjected to EQ. Those of us who use home-theater preamplifier-processors and audio/video receivers (AVRs) should be familiar with such constraints. Remember the hue and cry that went up when it was revealed that even top-of-the-line processors downsampled everything to 48kHz in order to use Audyssey EQ? Nothing has changed—increasing DSP horsepower to enable higher sample rates costs money for something that the vast majority of buyers don't care or even know about.

In the more specialized market for music servers, it still takes pricey and heat-generating DSP horsepower to handle higher sample rates. Multiply everything by three to take the step from two to 5.1 channels—not to mention 7.1 channels—and makers of proprietary servers have logically decided that the added cost of accommodating those few multichannel-music fans would create a significant cost deterrent for the bulk of their market. I do get it.

The result is that the server market is dominated by a wide range of products that take no notice of music files of more than two channels. A very few of these give a slight nod to multichannel by supporting multichannel files, but unless they take on the cost of including beefier processors, that support is limited. Usually, such servers will play files of resolutions up to 24-bit/192kHz PCM or DSD64, yet don't comfortably support DSP. I've reviewed a number of these, and appreciate them for offering multichannel at all. After all, there's no reason any stereo music server with digital input from internal or external storage and digital output via USB, I2S, or HDMI can't be a multichannel music server—unless its CPU is a complete wimp.

At the other end of the market are the hands-on folks who configure and build servers based on the Windows or Linux operating systems. (Due to Apple's design policies, there are significantly fewer MacOS/iOS examples.) These DIYers take special care with every component, from power supply to input and output devices to CPU, RAM, and storage, to ensure that the result generates the absolute minimum of acoustical and electrical noise, and is isolated from external sources of such noise. I take my hat off to them, and might join them one day.

Right now, my comfort spot is in the middle. In my January 2019 column I reviewed Wolf Audio's Alpha 3 server, which, like the Baetis Prodigy X server I own and love, is what a sophisticated DIY music lover would build if she or he had access to the OEM source market. Each is basically a PC running Windows 10 that employs a powerful Intel i7 CPU in a fanless case, to ensure that it runs silently enough for a music room. The CPU in these servers is substantially faster and more capable than those in proprietary stereo servers, and that's what makes all the difference. There are many other products that are the results of a similar approach to design and construction, and until recently they were the only option for committed multichannel listeners.

Roon Labs Nucleus+ music server
It was reasonable to expect that the philosophy that led Roon Labs to support multichannel playback and DSP/EQ might lead to a complementary physical product, and the release of the Roon Optimized Core Kit (ROCK) buoyed my hopes. Roon describes Roon OS as a "a custom Linux based operating system, tailored for running Roon Server and providing a best-in-class, appliance-type user experience to host the Roon Core," and ROCK as "a do-it-yourself build of Roon OS." They recommend that ROCK be installed only in an Intel NUC. Such customization of the OS for a single application relieves the CPU of the burden of all the non-Roon, non-audio functions that constantly belabor CPUs running a general-purpose Linux or Windows OS, and would greatly enhance the CPU's ability to process audio files. Buying a NUC and installing ROCK isn't difficult, but the prospect of selecting and dealing with a bare-bones box intimidates many.

With the Nucleus+ ($2498) and the less powerful Nucleus ($1398), Roon has not only made an appealing, ready-to-wear server, but has optimized the idea behind the ROCK kit (footnote 1). Both Nucleus servers are based on an Intel NUC board, but in a small, silent, elegant package, and their operating software is an even more specialized version of ROCK. This lean, mean version eliminates support for CPUs other than Intel's i3/i5/i7, any peripherals not included in the Nucleus, and a cooling fan and its control circuitry.

When I removed the Nucleus+ from its box, I was surprised by its small size and solidity—at 8.3" wide by 2.95" high by 6.1" deep and weighing 5.5 lb, it looks and feels like a sturdy external power supply. Its top and sides comprise ranks of heatsink fins, its front panel is devoid of anything but the Roon logo, and all connections are in a small bay on the rear panel: two USB 3.0 ports, Ethernet (RJ45) and HDMI jacks, and a barrel power connector for the 19V wall wart power supply. The only actual control is an on/off button, and above that is the only status indicator: a soft blue LED. There's no WiFi. Very basic and very compact, the Nucleus+ is entirely suitable for a network appliance that does its job at the behest of control devices such as tablets, smartphones, or other computers, and gets its data from such storage devices as USB or NAS drives. (A hard drive can be installed for internal storage.)


The Nucleus+ uses an i7 processor, though Roon doesn't identify the specific model. I removed the bottom plate and opened the case to get a look, but couldn't—the CPU is thermally bonded to the underside of the top panel, and directly accessing it would have required major surgery. Even then, it wasn't clear that the chip's identifying label would be visible. On the one hand, this is frustrating—i7s range widely in their processing capability. On the other, there's no performance reference for an i7 in a comparably streamlined environment such as the Nucleus+, though I do know what works for me in a very different Win10 environment. The only real issue was whether the performance would justify Roon's design choices.

Because the Nucleus+ does only one thing, installing it was easy: I plugged in the power supply, connected an Ethernet cable from my LAN switch, and pushed the power button once. I used as the control device a MacBook Pro that sits on my equipment rack, and as soon as it found the Nucleus+ I saw the familiar Roon greeting asking me to "Choose your Core" (Nucleus+, of course), "Sign Up or Login" (I did), and "Where do you keep your music?" (on my NAS). When I returned from lunch, Roon had loaded my more than 65,000 files into its library and had begun analyzing them. That process would continue, in the background, for about a week, but otherwise the music player was ready to go. All I needed to do was to choose an audio device from the list Roon had found on my network. I chose the exaSound PlayPoint Network Audio Player connected to my exaSound e38 DAC. That was it.

Because of its now-familiar, ultra-discrete mix, I always play Willie Nelson's Night and Day (DVD-Audio, SurroundedBy Entertainment SBE1001-9) with a new setup. It confirmed that all 5.1 channels were being played through the appropriate speakers, and that I was enjoying clean, well-balanced sound from each. After that, I can't say I was inspired to dig through my collection in search of revelations. That's because I firmly believe that, unless I'm using some kind of up/downsampling, EQ, or other manipulation, or have changed the type of connection hardware (USB, S/PDIF, HDMI, etc.), music players/servers do not contribute to the character of the sound.

To reconfirm this belief, I made a number of comparisons within the constraints of the equipment on hand. There is no Linux support for direct USB connection to my exaSound e38 DAC and there's no support in Roon for the miniDSP U-DIO8 multichannel interface that connects to my three Benchmark DAC3s. Using the Roon app, I compared the sound of the Nucleus+ to that of the Baetis Prodigy X, and also to both the Baetis server and the exaSound PlayPoint server using JRiver Media Center, in all cases directing output to my exaSound e38 DAC. After that, I compared files played through the Benchmark DAC trio with the U-DIO8 connected via USB to the MacBook Pro used to control Roon and the Nucleus+ with the same devices connected via USB to the Baetis. I concluded that any differences I heard in each comparison were so minuscule and fleeting as to be inconsequential. I also ran it via HDMI with the Marantz AVR in my office and the combination was faultless.

The only remaining question was how the Nucleus+ supported DSP. The answer: It depended on the sample rate and the DSP operations used. In the Roon app, if you click on the little star to the right of the track name in the playback panel at the bottom, you'll pop open a window with playback information. Near the top is the device name, and below it a list of all the steps in Roon's signal path. If you see a line of information between these two items, it numerically indicates how much headroom there is in the CPU's processing ability. The number references the total processing power of the system: "1x" indicates that it's running at full capacity to keep up; larger numbers (eg, "2x") indicate that the system has more capability than required. Numbers below "1x" mean that the system is incapable of performing the task in real time and are generally associated with interruptions in playback. If there's no such line of information, the Nucleus+'s processor is just loafing along with no stress at all.

Footnote 1: Roon Labs and Roon community.

Archimago's picture

Always great to hear about multichannel capabilities of software and techniques like the Auro-Matic 3D. I'm sure like any automated DSP "spatialization" technique, there are limits. Some nice demo tracks, some not-so-nice.

Curious if there is a way to let us know what i7 processor is in the Nucleus+? This will help put into context the processing demands of the Roon DSP for those looking at building their own machines while using the Nucleus+ as a bit of a benchmark for comparison...

Kal Rubinson's picture

As I indicated, the i7 device in the Nucleus is not accessible as it is, apparently, glued to the heatsink on its identifying surface. I tried.
Roon will not say.

misterc59's picture

It is obvious from you response, that since the i7 device from y our review sample is glued with the pertenant side facing away from the consumer, that (perhaps for a perceived economic advantage) Roon wants to keep this a temporarary secret. obviously, any competitor can take things part just out of "curiosity" and see which I7 device is being employed. It all comes out in the wash, reverse engineering and all that, but of course, it would be nice to know now.


misterc59's picture

Is there a law saying the component in question couldn't have the information stamped on the part in question, scratched out (or otherwise) to prevent others from identifying it just by visual assessment? I'd suspect this is illegal, but I'm not familiar with this territory.


Kal Rubinson's picture

I did open the box and poke around but going any further would have been destructive.

Archimago's picture

Yeah, can't imagine what the advantage would be. After all, the company to their credit provides ROCK for anyone who wants to install it on a NUC so it's not like they're dissuading anyone from building their own hardware.

It's just nice to know what speed the company's own devices run at to put Kal's results in context and to price out custom builds with fanless enclosures.

Kal Rubinson's picture

Archimago wrote: Yeah, can't imagine what the advantage would be.

The advantage to Roon is that they can get good thermal transfer from the i7 to the chassis/heatsink without bulky and expensive heat pipes or other devices. It seems a reasonable approach dictated by cost/size issues and not an obvious effort to hide anything. Now, their unwillingness to tell us the identity of the CPU is another issue. :-)

The Tinkerer's picture

And it states the Intel NUC "NUC7i7BNB". The CPU associated with this is the Intel® Core™ i7-7567U. This is a 2c/4T i7 with a 3.5ghz base frequency and 4.0ghz boost frequency across both cores. Fairly stout with most modern instruction sets employed as I type this in July 2022.

I purchased my Nucleus+ used and upgraded to 16GBx2 of Crucial dual rank memory (DR being proven to reduce CPU load over the same amount of SR/single rank RAM). Going with two SODIMMs over the original single module also means doubling the memory bandwidth. This resulted in a meaningful increase in performance for upsampling two channel PCM from Qobuz (48/24 -> 384/24); 21x to 23.5x. However, I also removed the original thermal paste (that was relatively dry) and replaced it with ThermalGrizzly Kryonaut Extreme. This resulted in that same process moving from 23.5x to /never/ going below 26x, regardless of ambient temperature. It usually hovers between 26.5x and 27.4x, depending on the song. Playing 192/24 without DSP or upsampling via Qobuz or internal storage to both my endpoints at the same time results in so much excess processing power that Roon fails to show a 0.0x rating in the signal chain at all.

Note that I have always used Auto Volume Leveling and this was employed for all above scenarios.

olderroust's picture

the roon boxes run Linux, no? if there is a way to issue commands at a shell, seems like the output of
less /proc/cpuinfo

ought to provide the information on cpu without need for a crowbar....

Kal Rubinson's picture

I agree. That should be possible except (1) it runs a custom Linux-based OS developed by Roon and (2) I have no way to do it.

mememe2's picture

With an average playing time of 4 minutes per file - my conservative estimate, that comes to 4333 hours of music. At 4 hours of uninterrupted listening per day that would entail 1083 days of listening. That would equal 3 years of consecutive nonstop listening for four hours per day. And that's only for a single listening of each file. Who has the time for that? Quantity vs. quality. I have a hard time believing that all of those 65,000 files are worth keeping or even downloading. Less is more.

Kal Rubinson's picture

I think the average is much higher than 4 minutes per file since the majority of my collection is classical. JRiver estimates it as about 6000 hours. Most are, indeed, are of high musical and audio qualities, both the classical and the non-classical.

That said, I have no hope or concern about listening to them all before I die or lose interest, regardless of which comes first. In fact, I am still adding new music to the collection. My motivation is to have what I want to hear whenever I want to hear it and I love chancing upon and rediscovering an old favorite. Would you deny that to me?

mememe2's picture

Of course not

Kal Rubinson's picture

Thanks. Logic has little to do with my passion.

Bertie Bucket's picture

Now imagine those songs were money. You could have more than you can spend in a lifetime so what's the point of that?