Sound Science

Neal Van Berg of Sound Science has been busy building his forthcoming flagship server, the Music Vault Epiphany (around $10k). As you can see, Eric has pulled out all of the steam punk stops with Epiphany but those looks are mostly functional. That Metropolis-looking tower is all heatsink, baby, to cool the Intel Core I7 Processor. There will be 32GB of RAM and 2TB of SSD storage as well as all of the sonically advantageous goodies Neal has garnered from building his Music Vault servers.

It's worth noting that Neal is an EE with 25 years of professional experience working in acoustics and vibration for Hewlett Packard, Bruel & Kjaer Instruments, and Agilent Technologies so some of the sonic goodies in the Music Vault servers come out of this experience and knowledge.

In use at RMAF 2015 was the Music Vault M7SS that comes with SSD storage ($6,995 w/2TB, $7,995 w/4TB, $9,995 8TB), 32GB of RAM, an Intel Core I7 processor, a number of outputs including AES/EBU, USB, WordClock (in/out), and HDMI. The M7SS also houses its own 24-bit/192kHz capable DAC with accompanying RCA and XLR output pairs. There's also an optical Blu-Ray drive for ripping (via dBpoweramp) and JRiver is the included player software of choice.

The Lyngdorf TDAI-2170 170wpc integrated amp/DAC (starts at $4,000) was handling everything but vinyl including digital room correction. The TDAI-2170's DAC supports up to 32-bit/384kHz PCM and DSD 128 playback.

A pair of Harbeth speakers filled out the system.

COMMENTS
tonyd's picture

Can be as simple as a PC running Linux + Logitechmediaserver + 2 external USB drives. A basic modern notebook works well. Costs start @ $600 and your time to set up.

Just saying ;-)

Michael Lavorgna's picture
...cost more.
VK's picture

I'm sorry, but this is totally overkill for no matter what type of home audio equipment. Unless if you're doing heavy stuff on-the-fly. And even in this case, i dont see the purpose of an top of the line enthusiast processor like an I7 and 32 gigs of RAM. Even some video workstations dont use more than 16 GB.

But of course, bigger numbers always look better in the marketing sheet...

Best regards!

Venere 2's picture

I don't feel completely qualified to judge this matter. But, I have the smallest (current) Mac Mini with 4 gigs of RAM. I play CD files and 24 192 files a plenty.

There is no stuttering, errors or bugs. It also sounds great! I would rather spend the money I save on better amplification or a better DAC (not that I am unhappy with what I have, far from it) than a supercomputer.

VK's picture

But i dont think that save money for other upgrades is the point. Even if this beast of server only costs US$ 1000, the specs are ridiculosly high! You really dont need that much.
But as always, with audio things are different, and you NEED the very top best! And it looks cool marketing wise...

Best regards!

Michael Lavorgna's picture
What would you say are the ideal specs for every music server?
whoozwah's picture

It's a bit of a loaded question because every use case will be different. A more relevant question to this thread might be can we come up with a function that might take advantage of the hardware put into the server that's on display here?

Possibly doing room correction, using it as a DAW for creating music rather than just listening. I can't really come up with too much else that would push the hardware that hard. Maybe someone else can :)

Just to serve up a counter example to the server here, my own system consists of a Raspberry Pi, Hifiberry DAC (which uses a burr brown and has an I2S interface), RuneAudio OS (free) and a USB2 HDD for serving my flac files. Whole thing costed under 200 bucks and sounds fantastic to my ears.

I don't think there's anything wrong with having a computer this powerful to play back digital audio. I do however think that if someone were to use it only for that purpose then it becomes a bit wasteful. There's so much you can do with that much computing power and I'm confident that the folks that end up buying this bad boy know it and will use it appropriately.

Michael Lavorgna's picture
I use an ordinary MacBook Pro and it sounds great. Every purpose-built music server and most streamers I've heard here sound much better.

I will be reviewing one of the Music Vault servers so we'll see if their approach makes sense sound-wise.

VK's picture

My point is that this music server is totally overkill. If an Core I7 with 32 gigs of RAM is the best thing, i can say that the super stable and all mighty Xeon E7 8890 v3 (18 cores and 36 threads) with the maximum amount of 1536 gigs of RAM will be the ultimate music server, and we know that it's not true, because it depends of other factors like how optimised are our OS and audio players.

This music server is like an set of speakers with more than 100dB/W/1m hooked at an 10 kW amp in an small room. You will never use that much power!

Mr. Van Berg will tweak the USB out ports, and the MB audio for better sound? This is ok, and i agree with this. But the use of such processor and RAM is beyond the actual needs of every audiophile. Just look at the exaggerated high end top C.A.P.S. from Computer Audiophile. Even their suggestions don't use more than a Core I5 or an entry-level Xeon with an already woopping 16 GB of RAM!

Please don't get me wrong. Like i said, unless if it will make use of heavy stuff on-the-fly (and i mean HEAVY), it doesn't make sense. Higher quality sound doesn't come from a highly clocked multi-threaded CPU and a large RAM.

But let's see how it will perform.

Best regards!

Michael Lavorgna's picture
...Mr. Van Berg has the freedom to pursue whatever avenues he feels will improve his products. We are free to buy them or not. The rest of this discussion is posturing, imo.
VK's picture

I agree! I hope to read a review of it in Audiostream in the future. Although i didn't agree with his approach, i'm always curious to read about how it performs ;)

Have a good weekend!

Venere 2's picture

Here is how you can raise the level of a lowly Mac Mini. Get an adapter and put one of these on it:

http://www.shunyata.com/our-products/pwr-cables/55-venom-series/415-veno...

One of the best upgrades I have ever done! Your Mac Mini will sound much better, and still cost less than some of these very expensive alternatives.

Jorgen Skadhauge's picture

So how did it sound ? with/without roomcorrection ?

Michael Lavorgna's picture
...with and without room correction.
marce's picture

No shielding, open to dust, wont conform to CE/FCC EMC regulations, heatsink fins wrong orientation for efficient air flow... Total joke of a design...

marce's picture

Oh forgot, more RAM more switching noise... SSN simultaneous switching noise...

Clever Dean's picture

That's the question I have. I have a Virtual Server HA (High Availability) Cluster that I built using commodity 4 core AMD processors, mirrored 480GB SSD's from SANDISK, 16GB RAM, Dual GB Ethernet in a LAG.

I'm able to run 6 VM's on it and with the RAM and SSD it runs like a bat out of hell.

The point I'm making is what in the AV world is going to make a $10,000 server even break a sweat.

You can run small companies with $10K in server hardware. We are talking 100's of employees.

Michael Lavorgna's picture
...the $6,000 and change servers I have reviewed have delivered better sound as compared to a MacBook Pro/NAS. Better sound is to my mind the problem we are all trying solve.
John Sully's picture

This machine is roughly equivalent to the workstation I run for development at work. That machine typically runs a couple of instances of Visual Studio, SQL Server, IIS, software for control of a Campbell Scientific data logger, Chrome with about 3 zillion tabs open, IE handling a streaming radio station, Outlook, yadayadayada. Sometimes I run VM's because I have to test code running in different environments. It still doesn't breath hard. Mine has an 8-core 3.5GHz Xeon and cost around $3K with a fairly ridiculous 3D graphics card.

In the early 1990's I worked at Silicon Graphics and we had a low end computer I worked on called the Indy which ran on a 100Mhz MIPS R4000 (and some up market chips in later versions) which could handle video capture/editing/playback in SD of course. It was the first workstation to come with a video camera. Somebody put up an antenna on the roof and digitized FM stations from the gulch and put them on the network. It was able to decode music and play it just fine (I had a pair of Cambridge Soundworks cube speakers/bass module which was about as good as you could get for a computer in those days) and it sounded fine. My favorite thing about the playback software? The undocumented --spinal-tap option, which changed all the controls on the control panel to go up to 11 (I always suspected that my brother put that one in). Someone else did it for video, too, you could watch TV over the company network, I used it during the OJ chase!.

The point of all this is that the job you are asking a dedicated music server to do is not all that demanding in terms of either CPU or I/O. Even 192/24 is only about 11Mb/s, which is nothing on busses whose speeds are measured in Gb/s. The guy who is using a Raspberry Pi has the right idea: you need to scale your resources to the job. Your music server is going to be on 24/7, at least mine is because I use it as a network server and I wake up at odd times of the night, so I want it to be as efficient as possible. It is easy to throw hardware at a problem, it is not so easy to thoroughly understand and analyze the computational and i/o demands which a system will have and size the hardware appropriately.

It does look cool. Especially if it has red led case lighting. :-)

Michael Lavorgna's picture
For example, the idea that the guy who is using the Raspberry Pi has the right idea, is a compelling idea. Reality, however, is another matter.

Do you make anything we can listen through, John? Or are your ideas just theories? I know the answer so don't bother answering. My point being, while ideas are lovely things, they don't amount to a hill of beans in this crazy, mixed up world. People Like Neal Van Berg's ideas on the other hand matter because he actually makes things we can listen through.

Arm-chair audiophiles who espouse expertise in things they don't have the will or nerve to implement strike me as so much...fluff.

John Sully's picture

Nope, audio was not my area of expertise. That was my brother's department. I worked on VM/IO/cache stuff at the interface between software and the CPU/memory system. I wrote the code which made the operating system work and worked with hardware designers to optimize these areas of the system.

My expertise would lie in understanding the demands on the system in terms of computational and i/o resources. And as I said above, audio really does not make a modern computer such as a Raspberry Pi or a Chromebox or the little Synology DS213j NAS I use as a server breath hard. Look, decoding a lossless file requires solving a polynomial over and over again using a statistical model developed during the encoding process. This is a piece of cake for a modern processor with floating point and a major selling point for FLAC and ALAC. If you like wav or aiff, the processor only has to program the DMA system to transfer the data to a buffer used by the USB controller or which is used by the network stack. Basically nothing. Do you understand why an i7 might be overkill for something like this? Or why 32Gb of memory is maybe a wee bit too much?

Just throwing more computational power at a computational problem does not give you better results. As long as you can keep up with the demands of the i/o device, whether it is network or USB the computing platform is adequate to feed the beast. I am not talking about DAC's or anything in that part of the chain, only about what it takes to keep the chain fed.

Michael Lavorgna's picture
This is a show report where I report on what I see and hear (sometimes) at the show with a focus on new products.

I love the "Do you understand..." bit though. Do you understand that I get your point but don't care?

Sound Science's picture

Michael,
I want to thank you for the great pictures and description of the gear on display in my room at Rocky Mountain Audio Fest 2015. I also wanted to reply to some of the comments from your enthusiastic readers.

It’s true that a Music Server can be as simple as a PC running Linux, a Raspberry Pi, or even an iPod feeding a surround sound receiver, but if you're not a "do it yourselfer," and have a desire for truly superior sound, the Music Vault is an excellent option. My customers want the very best uncompromised performance and Music Vaults have been delivering that for longer than any other commercially available Music Server. Numerous customers have traded up to the Music Vault, selling their previous music server because they felt the Music Vault sounded better and was easier to use.

Then there's the newly designed Music Vault Epiphany. It's a prototype that was on static display in my room at RMAF, which when it's finished I hope Michael will review it. The heat sink in the Epiphany that was brought into question was designed to conservatively dissipate the thermal energy produced by the I7 and is a standard product I purchase.

In the Epiphany I want customers to see what they're getting inside, and it was designed the way it was because we thought it looked cool. That heat sink alone costs 3 times what the Raspberry Pi costs. The digital and analog inputs and outputs cost more than any of the simple systems described in the comments. And you can hear those differences. If someone is happy with their server that's great, but if there's a desire for equipment that sounds superior and isn't a "do it yourself" project, an option is to check out the Music Vault line of Music Servers. I think my customers would tell you they're happy they bought their Music Vault. You can read many of their comments along with the Music Vault specs at my website www.musicvaultfcat.com

EuroDriver's picture

Hi Michael,

I would like to chime in on the discussion with overkill in computer hardware for a music server.

There are 2 distinct roads and philosophies for achieving good results in computer audio.

The first road is the minimalist, keep the digital processing to a minimum and deliver the bits to the DAC as cleanly as possible with the minimum of electrical noise. Software players which are excellent performers in this genre are MPD, MQA, JPlay, CMP2

However, a second way has become extremely interesting with the availability of very good DSD capable DAC's that can handle DSD128 and DSD256. With these DAC's one can use software running on PC's to convert Redbook and Hi Res files to DSD 128 and DSD256, realtime "on the fly"

The amount of processing involved for realtime PCM to DSD conversion is an order of magnitude greater than with the minimalist software players and requires a quadcore processor with a CPU benchmark of more than 6,000 to be able to process on the fly without stuttering during playback.

However, heavy processing in the PC also comes with its electrical noise baggage, and this needs to be mitigated in order to achieve the best sound quality results

Michael Lavorgna's picture
;-)

It's my experience that there are nearly as many roads as there are people. The important thing to keep in mind, at least in my mind, is we listen through products in systems not ideas about pieces of products.

I very much appreciate your thoughtful comments.

Cheers.

Archimago's picture

Seriously, let's not overstate the need for computing power.

Realtime transcoding to DSD128 in JRiver can be done with a reasonable, cheap Pentium G3220 CPU which is what I use. Underclocked even and FANLESS for silence. No problems with realtime stuttering or dropouts. No need for quad core. DSD256 might even be achievable with this CPU. Recognize that higher DSD sampling does not necessarily achieve better subjective sonics.

If you want to put CPU processing to good use, do digital room corrections. Even with a 24/192 correction/convolution filter, the fanless G3220 CPU has no problems keeping up.

The computer science guys above are right. Music player/servers like this doesn't make sense... Too expensive, wasteful of money that can be used for other purchases (like good music), and wasteful of electricity.

EuroDriver's picture

Hi Archimago,

I enjoy reading your posts, and appreciate the effort you put into your analysis and reports.

Out of the three software players that do PCM to DSD conversion on the fly, HQ Player has a multitude of options for choice of filter, dithering, and sigma delta modulation. There are further settings in the software that can distribute the processing over multiple cores. Unsurprising this feature improves perceived sound quality on a quad core CPU, and degrades perceived sound when used on a duo core machine. One filter which can sound very good is the Polysinc filter, but it takes about 35 seconds to set up when running on a i7 3770. A faster more powerful processor reduces this setup time. The sigma delta modulators which are user selectable include 5th order and 7th order and have there own sound signatures which also change according to the processor characteristics.

With HQ player's multitude of settings there is no one CPU size that fit's all

Your point about the benefits of using DSP to do room correction are well taken. In an acoustically challenged listening space there I would agree that there is more mileage to be gained from digital room correction, but for listening rooms with advantageous dimensional ratios, and appropriately acoustically treated, then the path of native DSD playback and PCM to DSD realtime conversion using software which does heavy processing to improve the quality of the sigma delta modulation is an avenue which can be exciting to pursue.

It's not for everyone, but for those who like to get goosebumps when listening to their favorite recordings, go for it !

prerich45's picture

I won't purchase one because I like to build my own stuff, but I'm excited and intrigued by the upcoming review. I like seeing stuff like this! :)

prerich45's picture

I went home and tested my PC (only 6 cores, 16gb ram, 12tb NAS) to see if it could successfully convert PCM (.WAV, .APE, .FLAC etc) to x2 DSD on the fly (thanks Eurodriver :) ). I've read of other people that have problems with that....no problems at all here. No stuttering, latency issues, nothing at all.

You may say why do you want to do that.. just do unmolested bitperfect. Why did the chicken cross the road? I think that's what Michael is getting at - someone will like this product and find uses for that type of processing power.

EuroDriver's picture

The DSD format has a lot of disadvantages compared to PCM, but in the playback context it does have a very big advantage in that the filtering that needs to be done is at a much higher frequency and can be done with a much gentler sloped filter than the filter that is needed for PCM of a similar data rate.

Gentle slope filters are less detrimental to perceived sound quality than steeper filters. The competition between steep PCM filters and gentler DSD filters is on going, but gentler DSD filters are costing less than the really good steep PCM filters at the present time.

X