DIY Server vs Off the Shelf (Innuos, Nucleus+, etc.)

[jonathan_s]

Hi All,

 

I have an old, not heavily modified Windows audio PC I built about 6-7 years ago (passively cooled, Paul Pang USB card) and am going to take the plunge to build a new server.  So I've been trawling through some of the massive threads on this particular forum to try to catch up with the technology. 

 

One thing I was curious about is whether any head to head comparisons had been done between a SOA one or two box DIY server and an off the shelf product like a Nucleus+ or Innuos.  As it stands, I'm hoping to build a one box solution using an AMD gaming motherboard, industrial ECC RAM, a high powered Ryzen processor, JCAT NET and USB cards, Euphony OS and the like.  Coupled with high quality linear power supplies, that is definitely in the neighborhood of a Zenith MkIII, so I'm curious if anyone can comment on the relative performance of a DIY effort as compared to an off the shelf unit.

 

 

 

 

[Nenon]

I think you are on the right track with your build. From there it all depends on the power supply. If you add a good state of the art power supply, you would surpass anything at the Zenith MKIII class and would be competitive with the Innuos Statement and Pink Faun. But again, at that point, it would be all about the quality of the power supply. 

Innuos Statement sounds really really good. And you would not have their custom OCXO clocks. But our findings here are that a faster CPU (to stress this again - with the right power supply!) sounds more dynamic than a low-noise CPU like the one Innuos uses. But keep in mind that the Statement is very well refined server and it would take a lot of tweaking to get even close to that level of performance with a DIY server. 

The Pink Faun servers on the other hand use a more powerful CPU and has even more OCXO clocks. Different philosophies, different approaches, and both end up with an excellent result, that is really really hard to reach with a DIY server. 

And then there is the Taiko Extreme, which is at whole different level and IMO unreacheable with a DIY server. 

But if you are aiming to reach or slightly exceed the performance of the Zenith MKIII you are on the right track. The JCAT USB card is very good. You can also consider the Pink Faun USB bridge with ultraOCXO clock, but that's a lot more expensive. 

[jonathan_s]

Nenon, thanks for the reply...  I have been especially looking at your build (hence the specs I stated earlier!!!), as I want to build a one box solution to cut down on the ultimate cost.

 

Thanks for the qualitative assessment, too... I am looking to spend Zenith MkIII money, and like the idea of being able to replace portions of the build as technology changes or new knowledge is gained.  

 

 

[Hardinge]

Hi @jonathan_s i can’t add much to your quest yet but think we’re on the same path. I have a Zenith MkII running Roon and love it (especially in squeeze mode). My library is too big and i want to take it to work and then home on weekends so to get 8tb internal SSD and to utilise the recent movement here to high speed cpu’s i’m making my own. 

HDPlex is out of the H3 cases so i’ve gone the H1 which will be great to transport but i need a thin mini-itx mobo to be able to fit a pink faun usb card. Intel based mobos that fit that profile are all long in the tooth. All this to get to my question, which AMD and Ryzen hardware are you looking at? I haven’t seen that chat about that brand path (only intel). 

 

[jonathan_s]

I was looking at Nenon's build details, with regard to an ASUS ROG Strix X470-I mobo and a Ryzen 7 3700X CPU.  The motherboard has one PCI and two M.2 slots.  I'd run the JCAT or Pink Faun USB card in the PCI slot and the JCAT NET card and an optane drive (for the OS (Euphony)) in the M.2 slots.   An M.2 to PCI adapter would be required to run the NET card (that is available on E-Bay, as posted by Nenon, and elsewhere perhaps).

[Hardinge]

Thanks @jonathan_s.  What case are you looking at?

[Blackmorec]

10 hours ago, Nenon said:

I think you are on the right track with your build. From there it all depends on the power supply. If you add a good state of the art power supply, you would surpass anything at the Zenith MKIII class and would be competitive with the Innuos Statement and Pink Faun. But again, at that point, it would be all about the quality of the power supply. 

Innuos Statement sounds really really good. And you would not have their custom OCXO clocks. But our findings here are that a faster CPU (to stress this again - with the right power supply!) sounds more dynamic than a low-noise CPU like the one Innuos uses. But keep in mind that the Statement is very well refined server and it would take a lot of tweaking to get even close to that level of performance with a DIY server. 

The Pink Faun servers on the other hand use a more powerful CPU and has even more OCXO clocks. Different philosophies, different approaches, and both end up with an excellent result, that is really really hard to reach with a DIY server. 

And then there is the Taiko Extreme, which is at whole different level and IMO unreacheable with a DIY server. 

But if you are aiming to reach or slightly exceed the performance of the Zenith MKIII you are on the right track. The JCAT USB card is very good. You can also consider the Pink Faun USB bridge with ultraOCXO clock, but that's a lot more expensive. 

As I remember a post from Nuno, Innuos tried a fairly large number of CPUs, including the higher power units and the CPU they selected sounded the best. Bear in mind they are using a custom mother board and they wrote and optimised their ‘for audio’ operating system, so there’s no unnecessary overhead. Essentially a more powerful CPU would require more power, which would produce more heat and noise. The OS Innuos designed shuts down a lot of processes while playing music, such that it’s really only RAM related functions that’re running during playback. Other operating systems are way more resource intensive so I’m not sure an Innuos would benefit from a more powerful CPU in the same way that other OSs like Windows would.  One of the reasons Innuos sounds so good in the first place is its PSs, another major reason is its innuOS operating system and low EMI motherboard. 

[StreamFidelity]

10 hours ago, jonathan_s said:

I'd run the JCAT or Pink Faun USB card in the PCI slot and the JCAT NET card and an optane drive (for the OS (Euphony)) in the M.2 slots. 

 

My recommendation is to use a motherboard and a chassis that allow the JCAT USB and NET cards to be connected without an adapter. I use for it

 

ASUS ROG Maximus XI Gene

and

Streacom FC9 Alpha (Black) - Desktop - micro ATX

 

Look left side on pic:

 

[Nenon]

8 hours ago, Blackmorec said:

As I remember a post from Nuno, Innuos tried a fairly large number of CPUs, including the higher power units and the CPU they selected sounded the best. Bear in mind they are using a custom mother board and they wrote and optimised their ‘for audio’ operating system, so there’s no unnecessary overhead. Essentially a more powerful CPU would require more power, which would produce more heat and noise. The OS Innuos designed shuts down a lot of processes while playing music, such that it’s really only RAM related functions that’re running during playback. Other operating systems are way more resource intensive so I’m not sure an Innuos would benefit from a more powerful CPU in the same way that other OSs like Windows would.  One of the reasons Innuos sounds so good in the first place is its PSs, another major reason is its innuOS operating system and low EMI motherboard. 

 

Innuos thinks a low powered CPU is the holy grail. Taiko thinks a dual Xeon CPU with 40 cores and 48 Gb of RAM is the way to go. 

 

I haven't tried the Statement in my system, but I have tried the Zenith MK3. And I also have heard multiple demos showing the difference between the Zenith MK3 and the Statement. Big difference by the way. The Statement sounds really really good. No idea how it compares to my DIY server, because I haven't had the Statement in my system. But if I can judge by the Zenith MK3 vs. Statement demos at the shows, I strongly believe I hear a similar difference (or more) when I compare the Zenith MK3 with my DIY server in my system. Does that mean my server is at the level of the Statement? I don't know!

 

@jonathan_s stated he would be using Euphony. That's a highly optimised operating system for audio where all the necessary processes are disabled. My findings (and not only mine) are that the faster the CPU, the better it sounds. An example - with everything else being equal, switching from a AMD Ryzen 3 2200G CPU (utilized at 0%-1% during play) to the much more powerful AMD Ryzen 7 3700X CPU (also utilized at  0%-1% during play) was a big uptick in sound quality. 

 

So, although I once was a fan of the Innuos philosophy, my vision is different now. I aim at processors with much higher performance, more cache and invest as much as I possibly can in the power supply. Don't know why but it works better in my system. 

[Nenon]

7 hours ago, StreamFidelity said:

 

My recommendation is to use a motherboard and a chassis that allow the JCAT USB and NET cards to be connected without an adapter. I use for it

 

ASUS ROG Maximus XI Gene

and

Streacom FC9 Alpha (Black) - Desktop - micro ATX

 

Look left side on pic:

 

 

I absolutely agree with you! That would be ideal.

 

However, I am still on the hunt for a good sounding AMD-based motherboard that is smaller than full size ATX, has at least two PCIe slots, and is fully passive (i.e. no active cooling of the chipset). I thought the AsROCK X470D4U would be perfect for that. But it did not sound as good as the smaller ASUS gaming motherboard I had. You can read more about that in my post below:

On 9/17/2019 at 9:38 PM, Nenon said:

Now that everyone's attention is on the EtherREGEN thread ( @Superdad announced the first listening impressions and final price) I can report my latest findings.

 

About a month ago I decided to compare two AMD-based motherboards - a server motherboard - AsROCK X470D4U - and a gaming motherboard - ASUS ROG Strix X470-I. Not only that, but I also wanted to find out if using ECC memory makes a difference. Here is my original post:

 

I was using both motherboards in the same configuration:

- AMD Ryzen 7 3700X CPU

- Sean Jacobs DC3 4-rail LPS (ATX - 12V, 5V,3.3V; EPS - 12V; all separate rails)

- PinkFaun I2S Bridge with OCXO clock

- Euphony / Stylus one box solution

 

I started with the ASUS ROG Strix X470-I motherboard and 2 x 8GB DDR4-3200 CL16 Non-ECC RAM. I had it for 3 weeks in my system. Just enough to get used to the sound. At that time I did a quick comparison between the AMD/ASUS and an AsROCK / Intel i9-9900K. Here is my post about that comparison:

 

Then I switched to the AsROCK X470D4U, keeping the same Non-ECC RAM. I did not like it from day 1, but I kept listening to it for almost 3 weeks. To be honest, I could not wait to go back to the ASUS. Each motherboard had its own sound signature, and I certainly prefered the gaming ASUS motherboard over the server AsROCK. There was no enjoyment with the server AsROCK. I could listen to background music, but that was it. 

 

But what about the ECC memory? Per the ASUS ROG Strix X470-I specs, it does NOT support ECC memory. The AsROCK X470D4U does. 

I did some research and decided to try the industrial wide temperature Apacer ECC RAM (https://industrial.apacer.com/en-ww/DRAM/DDR4-Wide-Temp.-ECC-UDIMM ). The idea was not that I needed usability from -40°C to +85°C, but that this type of RAM was built with better quality parts and may sound better. Plus, I've heard from a few places that ECC RAM sounds better. The Taiko Extreme server uses ECC RAM for example. JCAT is also a big fan of the Apacer ECC RAM. 

 

I installed the Apacer RAM on the AsROCK motherboard and heard an improvement immediately. I still think I prefered the ASUS motherboard, but they got close. I wish I could install the Apacer ECC RAM on the ASUS motherboard. Well, I had the suspicious that although the ASUS ROG Strix X470-I motherboard does not officially support ECC memory, it may actually work. So I gave it a try. AND IT WORKED! I was able to boot the ASUS ROG Strix X470-I with the Apacer ECC memory. And it sounded really really good. I could not stop listening music all night that day. And I am still enjoying it :).

 

I still have the Intel i9-9900K / AsROCK Z390 Phantom Gaming-ITX/ac. So I decided to do another quick comparison between the AMD/ASUS and Intel/AsROCK gaming motherboards. Unfortunately, the PinkFaun I2S card does not work with my Intel CPU, so I had to switch to USB. I used the motherboard USB output to compare. That's the same comparison I did in the beginning of August (see above). The only difference was that this time I used the ECC RAM on the AMD/ASUS. And while the soundstage collapsed, because I was using the motherboard USB output, not the PinkFaun with ultra OCXO, I do believe I liked the AMD better this time. 

 

That leads me to a conclusion that the RAM, a component I always neglected in the past, is very important. That industrial Apacer ECC RAM is really good. I don't know if it sounds so much better because it is good quality or because it is ECC. My guess is that both contribute to what I am hearing. 

 

To summarize for those who did not want to read the entire long post:

- The gaming ASUS ROG Strix X470-I motherboard sounded a lot better than the server AsROCK X470D4U motherboard (to my ears, in my system).

- The industrial Apacer ECC RAM made a big difference. 

- It just happened that although not officially supported, the ASUS ROG Strix X470-I motherboard works fine with the Apacer ECC RAM, so that's what I ended up using. 

 

Note: I am assuming that everyone reading this post understands the difference between ECC and non-ECC memory.

Intel purposely does not add support for ECC memory to their consumer grade CPUs such as the Core i7 and Core i9, so they cannot be used with ECC memory. That option is reserved for Intel's server-based CPUs (i.e. Intel Xeon), which typically require server-grade motherboards. So Intel-based gaming motherboards typically only work with consumer CPUs and don't support ECC memory.

AMD on the other hand does not have such restrictions. This is why we can use ECC memory on some consumer grade AMD CPUs, such as the AMD Ryzen 7 3700X in my case.

 

It turned out the ASUS motherboard with my I2S card on the PCIe slot and the JCAT NET Femto card connected via a M.2 to PCIe extender sounded better. I used one of these:

 

The overall result was better than the sound I got from the AsROCK X470D4U motherboard with the JCAT NET Femto connected directly to a PCIe slot. But to improve things even more, I replaced the cable with 24AWG 5N pure silver wires with proper shielding. 

 

It is easier to find an Intel-based motherboard that meets all my requirements than AMD. But my PinkFaun I2S card with ultraOCXO clock does not work on consumer Intel CPUs, so that's my limiting factor. 

 

It's all about the overall result at the end, not about the individual components. 

 

[Blackmorec]

3 hours ago, Nenon said:

 

Innuos thinks a low powered CPU is the holy grail. Taiko thinks a dual Xeon CPU with 40 cores and 48 Gb of RAM is the way to go. 

 

I haven't tried the Statement in my system, but I have tried the Zenith MK3. And I also have heard multiple demos showing the difference between the Zenith MK3 and the Statement. Big difference by the way. The Statement sounds really really good. No idea how it compares to my DIY server, because I haven't had the Statement in my system. But if I can judge by the Zenith MK3 vs. Statement demos at the shows, I strongly believe I hear a similar difference (or more) when I compare the Zenith MK3 with my DIY server in my system. Does that mean my server is at the level of the Statement? I don't know!

 

@jonathan_s stated he would be using Euphony. That's a highly optimised operating system for audio where all the necessary processes are disabled. My findings (and not only mine) are that the faster the CPU, the better it sounds. An example - with everything else being equal, switching from a AMD Ryzen 3 2200G CPU (utilized at 0%-1% during play) to the much more powerful AMD Ryzen 7 3700X CPU (also utilized at  0%-1% during play) was a big uptick in sound quality. 

 

So, although I once was a fan of the Innuos philosophy, my vision is different now. I aim at processors with much higher performance, more cache and invest as much as I possibly can in the power supply. Don't know why but it works better in my system. 

Well i hope your server does sound as good as a Statement.....

As for the rest, I’ll bow to your superior knowledge and experience....I’ve never  exchanged a CPU and am highly unlikely to ever do so.  However I will also be cautious about accepting that Innuos is no longer the right approach. Since I’ve had the Statement I’ve finished optimizing my network and added better umbilicals to the Statement and it has rewarded handsomely by continuing to sound even better, to the point one needs different adjectives to describe what the improvements sound like;  so I’ll hold judgement until I clearly hear something better. 

If more powerful CPUs do ultimately prove superior than I’m happy to look forward to even better sound in future. Progress. 

[Nenon]

1 minute ago, Blackmorec said:

Well i hope your server does sound as good as a Statement.....

As for the rest, I’ll bow to your superior knowledge and experience....I’ve never  exchanged a CPU and am highly unlikely to ever do so.  However I will also be cautious about accepting that Innuos is no longer the right approach. Since I’ve had the Statement I’ve finished optimizing my network and added better umbilicals to the Statement and it has rewarded handsomely by continuing to sound even better, to the point one needs different adjectives to describe what the improvements sound like;  so I’ll hold judgement until I clearly hear something better. 

If more powerful CPUs do ultimately prove superior than I’m happy to look forward to even better sound in future. Progress. 

 

Well, it is already there, although not at very favorable price point. Taiko Extreme. 

[StreamFidelity]

31 minutes ago, Blackmorec said:

If more powerful CPUs do ultimately prove superior than I’m happy to look forward to even better sound in future.

 

What is a higher CPU required for? For my case: Because of the algorithms! As far as I know, HQPlayer with the high-precision EC modulators for DSD require 4-6 cores with at least 4GHz. As much as I admire the Taiko Extreme, the lack of HQPlayer support (which used to be the case) is unfortunately no go for me.

[jonathan_s]

On 12/12/2019 at 5:13 AM, StreamFidelity said:

 

My recommendation is to use a motherboard and a chassis that allow the JCAT USB and NET cards to be connected without an adapter. I use for it

 

ASUS ROG Maximus XI Gene

and

Streacom FC9 Alpha (Black) - Desktop - micro ATX

 

Look left side on pic:

 

 

 

I can only hope my build looks this great when I'm finished!  Nice work!

[dminches]

My current build is in a Streacom FC-10.  Sounds like I should get an FC-9 so I can eliminate the adapters.

 

I assume the FC-9 supports the CPU riser?  I need to use that.

 

[cjf]

I would say that if you've got the knowledge to build your own there is literally no reason to buy a canned/pre-built off the shelf music server.  I suspect that market is for those without the experience, patience or motivation to build their own.

 

I have no doubt a properly setup DIY music server can and probably will surpass a pre-built unit since you can use all costs towards better parts where a pre-built unit has the cost of profits baked in thus something is going to be sacrificed somewhere in order for the creator to get paid.

 

Nothing wrong with either approach of course but me personally would never buy an off the shelf unit since I know how to build my own from scratch.

[Blackmorec]

4 hours ago, cjf said:

I would say that if you've got the knowledge to build your own there is literally no reason to buy a canned/pre-built off the shelf music server.  I suspect that market is for those without the experience, patience or motivation to build their own.

 

I have no doubt a properly setup DIY music server can and probably will surpass a pre-built unit since you can use all costs towards better parts where a pre-built unit has the cost of profits baked in thus something is going to be sacrificed somewhere in order for the creator to get paid.

 

Nothing wrong with either approach of course but me personally would never buy an off the shelf unit since I know how to build my own from scratch.

I follow the logic but I’m still no sure I buy in 100%. If the server manufacturers are using off-the-shelf components than sure, you’re logic is 100%. But as soon as manufacturers start writing their own ‘for audio’ operating systems and creating their own ‘for audio’ mother boards etc. the potential  to get something far superior to off-the-shelf PC hw and sw, which is all built for general computing, with all the extra overhead that entails must be rather high. 

[Summit]

8 hours ago, cjf said:

I would say that if you've got the knowledge to build your own there is literally no reason to buy a canned/pre-built off the shelf music server.  I suspect that market is for those without the experience, patience or motivation to build their own.

 

I have no doubt a properly setup DIY music server can and probably will surpass a pre-built unit since you can use all costs towards better parts where a pre-built unit has the cost of profits baked in thus something is going to be sacrificed somewhere in order for the creator to get paid.

 

Nothing wrong with either approach of course but me personally would never buy an off the shelf unit since I know how to build my own from scratch.

 

A good amp, DAC or server is more than the sum of its parts. To put together good parts can get you good sound and is often more inexpensive. To get excellent sound you will need more than good parts, you also have to test and select how the components work together and even modify some of the components.  

[dminches]

8 hours ago, cjf said:

I would say that if you've got the knowledge to build your own there is literally no reason to buy a canned/pre-built off the shelf music server.  I suspect that market is for those without the experience, patience or motivation to build their own.

 

I have no doubt a properly setup DIY music server can and probably will surpass a pre-built unit since you can use all costs towards better parts where a pre-built unit has the cost of profits baked in thus something is going to be sacrificed somewhere in order for the creator to get paid.

 

Nothing wrong with either approach of course but me personally would never buy an off the shelf unit since I know how to build my own from scratch.

 

I think you are ignoring the design aspect and assuming that all these builds are just a combination of parts.

 

I doubt I could ever come close to the taiko Extreme.  Could you?

 

[mourip]

18 hours ago, dminches said:

My current build is in a Streacom FC-10.  Sounds like I should get an FC-9 so I can eliminate the adapters.

 

I assume the FC-9 supports the CPU riser?  I need to use that.

 

 

I have used an FC5 for several years as my Roon Core server and for JRiver before that. It is an excellent roomy fanless case that has allowed me to make a lot of mods and to experiment but it has only one horizontal external PCI slot, making it necessary to use a PCIe riser cable. I do not know if the riser makes an audible difference with my JCAT FEMTO network card but if I had it to do over again I would definitely get the FC9 with its 3 vertical external slots. 

[Iving]

Can't report comparatively (with off-the-shelf), but the advantages of a DIY server include as already mentioned a case high enough for PCIe cards without a riser (mine is a Streacom F12C) - which can also permit certain no fan CPU cooling arrangements (e.g. NOFAN CR-80EH Copper IcePipe) and permit mobos spec. enough to have a sufficient number of "CPU-direct" lanes (which in my case includes one for an Optane PCIe AIC). Plenty of room for fitting an ATX LPSU too.

 

[cjf]

10 hours ago, Blackmorec said:

I follow the logic but I’m still no sure I buy in 100%. If the server manufacturers are using off-the-shelf components than sure, you’re logic is 100%. But as soon as manufacturers start writing their own ‘for audio’ operating systems and creating their own ‘for audio’ mother boards etc. the potential  to get something far superior to off-the-shelf PC hw and sw, which is all built for general computing, with all the extra overhead that entails must be rather high. 

What makes you so sure that these manufacturers are writing their own OS's? Is it because they are simply saying so? Also, how many of those type of systems that claim to have special sauce going on at the OS level allow the end User to even gain R/O access to see whats been done that is so different? My guess is none of them but maybe there are a handful.

[cjf]

6 hours ago, dminches said:

 

I think you are ignoring the design aspect and assuming that all these builds are just a combination of parts.

 

I doubt I could ever come close to the taiko Extreme.  Could you?

 

Wow, that "Taiko" certainly looks impressive. Lots of Copper, expensive milled aluminum, Dual Xeon CPU's, Dueland and Mundorf Caps..etc

To me, it looks like someone just found every expensive part that exists and crammed it all into one case to see who is foolish enough to buy it. In two years time when things start failing as they tend to do on computers you would have purchased the most expensive brick ever conceived. Kind of like that decked out Mac discussed in another thread at $52K.

So no, I would not come close to the Taiko because I know better than to build/buy such a device and think it actually has any real life use case except for someone with money to burn and no idea what they are buying.

You could spend $500 on room correction instead and have "Real" gains 10x over whatever that fancy box might claim to offer.

[Dutch]

53 minutes ago, cjf said:

except for someone with money to burn and no idea what they are buying

These people probably have a better idea of what they’re buying than you of what you are ‘burning down’ here. You seem to have just ‘met’ the product so you come across as a bit presumptuous.

Lots of information about the Taiko here with active contributions by the designer: https://www.whatsbestforum.com/threads/taiko-audio-sgm-extreme-the-crème-de-la-crème.27433/

 

55 minutes ago, cjf said:

You could spend $500 on room correction instead and have "Real" gains 10x over whatever that fancy box might claim to offer.

Once you get to that level where you don’t need DSP for room correction because your room is acoustically treated or purpose built this is no longer the case.

 

As always, garbage in=garbage out and from what I understand this is a ‘helluva’ source.
 

But yes, of course, it’s expensive without a doubt and for the average Joe a lot can be done and should be done first before even considering a product like this but there is a market for it where it’s not out of place. Everything’s relative. (And no worries, I too would never spend this much on a music server but for me that’s no reason to be this dismissive about a product.)

[Hardinge]

On 12/12/2019 at 10:13 PM, StreamFidelity said:

 

My recommendation is to use a motherboard and a chassis that allow the JCAT USB and NET cards to be connected without an adapter. I use for it

 

ASUS ROG Maximus XI Gene

and

Streacom FC9 Alpha (Black) - Desktop - micro ATX

 

Look left side on pic:

 

 

Thanks for the suggestions. The FC9 case seems a winner and the hdplex 800w internal unit also. Can i ask what CPU you have installed?