Building a DIY Music Server

[Foggie]

Aren't these comparisons of the extreme and similar a bit skewed?  99.9% of the people will never have / afford such a device. 

 

Which then begs the question, where does the hypothetical "custom server" price point (direct connect to DAC) start to where some believe it surpasses a well implemented endpoint?  Which even stating that is pretty wide to interpretation / many variables.

 

The endpoint vs non endpoint (or direct connect) discussion are like cable discussions in many ways. 

 

I am genuinely curious though and have toyed with the idea, just haven't found a conceivable way to test the "theory".  I have tried with the resources available to me (multiple high end wkst in so far as performance), but the endpoint has won out.  Can't really justify spending $5k (whatever that number is) just to see so to speak.

[Foggie]

17 minutes ago, matthias said:

 

You get the benefit without endpoint at much lower price points, please have a look at post #342.

 

Matt

Yeah I don't quite understand that.  An extremely well optimized (clocking, USB reg) optical endpoint using LPS, is bettered by a noisy computer albeit with a good USB output (jcat).  I certainly don't discount that in any way and every setup is going to be different - in most cases drastically different.  That's why like cables, there isn't any general rule that says method 1 is better then method 2.

 

My experience has been the opposite, although the USB output from server isn't "audiophile".  However, t is running 2019 SVR, W2135, custom MB, ECC mem, rtx graphics etc..(significant cost).  I suppose for a small layout one could try a dedicated USB card. 

 

I would love to get my hands on one of these builds (not crazy $$) and experiment for myself.  

[Foggie]

36 minutes ago, guiltyboxswapper said:

Agree.
 

 

 

I think there has been an advancement in recent tech; I was previously a endpoint type till JCAT XE + Ryzen 2900X came along.

Yeah for sure.  Sounds like a great setup you have

[Foggie]

2 hours ago, matthias said:

@Foggie

Here is a nice short summary from @romaz about "endpoints":

 

https://www.whatsbestforum.com/threads/taiko-audio-sgm-extreme-the-crème-de-la-crème.27433/page-99#post-616560

 

Matt

That's an interesting thread / info, thanks.  However, those guys are in the stratosphere $ and not apples to apples with many of us.  Even in that thread there are differences of opinion, as one would expect, on certain logic.  But as @guiltyboxswapper pointed out his setup was pretty straightforward, but I digress.

 

I don't stream and only use local library, which in of itself brings a slew of other variables (streaming).  Definitely some things to think about though. 

 

I could see trying out a dedicated USB card with LPS (like you pointed out) and putting the server in the listening room, direct to DAC.  I'll have to do some research. 

 

I was at this very same point when deciding on a OR, because I had tried some "general" direct to DAC tests and the endpoint setup was my preferred config, especially since I have no copper/eth, no converters and only running optical.  Fun stuff for sure.

 

[Foggie]

On 1/24/2021 at 11:35 PM, Nenon said:

 

Roon does not just serve the URL to an external program. It does a lot more than that that deteriorates the sound quality no matter where you run it or how you isolate it. 

Your idea to have Roon just handle the URL and shutdown any other activity is great. Unfortunately Roon has no desire to make such changes... or any changes that significantly improve the SQ. They have different priorities. Many of us have tried to convince them to do things to improve the sound quality but they have been ignorant and arrogant for the most part. That's their loss. New and much better software is coming up and replacing Roon in many high-end systems. The interface will slowly catch up over time.

 

Understanding the logic you stated, but are you indicating that the mere presence of roon is creating noise / crap on the network and ultimately degrading sound even if your using a well isolated endpoint / HQP?

 

 

[Foggie]

3 hours ago, Nenon said:

 

No, not at all. 

 

First, I don't really buy into the whole concept of a well isolated endpoint from the server. The idea sounds really good and logical but never worked well for me in practice. It's just like the concept that DAC manufacturers have been trying to convince us for years - the USB input is galvanically isolated and reclocked, so the quality of the server does not matter. No need to explain to the people following this thread that a well-designed source (server/endpoint/whatever you want to call it) makes a big difference. There is a really good reason servers like Innuos, Taiko, PinkFaun, Antipodes, etc. exist and cost so much. Arguably, they could make as much difference as a DAC (yes, even when connected to that same DAC with galvanically isolated USB input and reclocked signal). Anyone with a resolving enough system who has switched from a Mac MINI or a laptop to one of these well designed servers has experienced that. 

 

So, if a galvanically isolated USB input with a reclocker does not work, why do we think that we can "isolate" a server from the endpoint and make the endpoint immune to the server quality? From all my experiments, the conclusion has always been the same: "we can't". I've tried everything I could think of - fiber ethernet, fiber ethernet with state of the art clocks, fiber ethernet with state of the art clocks and state of art power supplies... and state of the art switches... and network cards.... and cables... and power conditioners... and state of the art motherboard clocks, powered by state of the art power supplies and cables... and different transceivers... and so on and so on. 

 

I don't know exactly why ethernet isolation does not work, just like I don't know exactly why a galvanically isolated and reclocked USB input on a DAC does not work as expected. But here are some examples that may give us a hint. Let's talk about an endpoint for a second - the last computer that is connected to the DAC. It may be receiving a data stream from the network (i.e. from a Roon Core) or playing music locally. 

1. In many cases, my experience with local music playback has been that disconnecting the network cable leads to a sound quality improvement shortly after the disconnect. That suggests that some noise is getting into the server from the network.

2. I have noticed that unnecessary network activity while playing music deteriorates the sound quality.

  2a. Euphony / Stylus caches the track you are playing and minimizes the network activity to a minimum. During some of my tests I noticed that opening multiple web browsers pointing to the Euphony page would degrade the sound quality. That was a very interesting experiment. The more web browser windows you open (pointing to Euphony), the easier it is to hear that. At some point I remember opening 50 windows on multiple computers. It turned out Euphony was using Ajax to update the time lapse interval display every second or so. That extra network traffic was enough to cause some sound quality degradation. BTW, I reported this to Euphony and they added an option to disable that traffic. 

  2b. When I isolated my endpoint to its own VLAN, I heard an improvement. Isolating to its own VLAN means that any broadcast traffic from my home network would not make it to my endpoint. I have a very small home network and have used a sniffer to see how much broadcast there is. It's very little, but yet enough to have an impact. 

  2c. I have tweaked how the NIC interacts with the CPU. You can configure the CPU interrupts and resources the NIC driver is utilizing, essentially how the NIC interacts with the OS. It's an interesting test. You configure it to use more resources and you get lower latency, but that comes with some side effects (i.e. your server becomes even more sensitive to any extra network traffic, thus making the network chatty Roon even worse). You make it use less resources, and you are more immune to all the network activity going on, but you increase the latency and that has other side effects. i will be writing more about this in the Windows LTSC guide later this year.

All that tells me that that network activity on the endpoint impacts the sound quality. Even if you have the perfect server streaming the perfect stream over the perfect network to your endpoint, the fact that the endpoint is receiving and processing network packets (while playing) is not ideal and degrades the sound quality of the stream going to your DAC. 

 

The ideal endpoint would be one that does not have a network connection at all. And this is where attention to detail in software that is designed with sound quality as top priority matters. I do expect to see some breakthrough innovations in the software in the near future. 

But if that is the case, why so many people buy into the server/streamer concept and after trying many solutions including a single server/endpoint device, they prefer it (in many cases by a big margin)? I think the answer is twofold. One part has to do with upsampling and the other with the quality of the server. Let's cover those two:

 

Upsampling:

Many DACs work better when they are fed by an upsampled signal. Those are typically DACs that upsample internally. By feeding them with an upsampled signal they have to do less processing. That results in less processing inside the DAC, which results in less noise, which ultimately means better sound quality. Sparing the noise that tha DAC is producing during its internal upsampling in some cases is much bigger improvement than upgrading servers/endpoints upstream of the DAC. That is of course DAC-dependant and not always the case. But for these DACs it makes perfect sense to use software like HQPlayer that does the heavy lifting and offloads the DACs extra processing (and noise).

What happens when you start doing heavy upsampling on your server that is directly connected to the DAC? Well, I mentioned above what the impact of the network/ethernet processing is, but that is nothing compared to the impact of the heavy processing that HQPlayer does. The noise generated from the upsampling process on the server is in orders of magnitude bigger than the negative effect of the network/ethernet processing. It's much better to isolate the upsampling noise from the endpoint and accept the network/ethernet noise / sound quality degradation. The net result is a big improvement in this case. And this is why many people prefer a two box solution - a server and a streamer. Sounds like a good compromise... but there are people like me who just can't settle with such a big compromise and kept looking. More on that later. 

 

Quality of the server (in a single box solution):

This is another aspect that we should not forget. When you have a single device that handles the server and the streamer part, it is directly connected to the DAC and its quality is crucial. Computers are noisy devices that were not designed for audio. But with a careful implementation and the right choice of software, hardware, power supply, etc. you can make a single box solution work better than anything else. You cache the tracks before playing, minimize the network activity, and take care of every detail, and you would be rewarded. 

 

There is a saying that your system is only as good as the weakest link in the system. The same applies to the digital source. I look at it as a complex chain of components that interact in a very complex way.  There are different solutions for different people. For some people the benefit of the upsampling is so big, that they can live with the negatives of a server and a streamer solution. Other people have DACs that don't benefit from upsampling and settle for the best single box solution they can get. There are also people who like the small / low powered endpoints and can't understand why we do all that instead of buying a *Rendu for example - from my experience those are people with small systems, typically with bookshelf speakers who don't need the massive scale, dynamics, and everything else a server like the Taiko Extreme can do but like the low noise, dark background, small but holographic soundstage of these devices (and I completely understand their point of view as well). There is no universal solution that works best for everyone. 

Here is what I do:

I currently use the Chord DAVE DAC powered by a Sean Jacobs DC4 LPS. This DAC benefits from upsampling. It specifically benefits from being fed by 705.6/768kHz PCM. It also benefits from more taps, but that's outside of the scope of this post. As mentioned above a two server solution does not work for me. So I upsample my local music offline and store it locally. There are programs like HQP Pro and some others that can do that. I store the music locally on my NVME storage. Then I use HQP and NAA on the same server. I assign affinities for those processes on different physical CPUs (which also means they use different RAM modules). I have a CPU that is only responsible for the music processes (i.e. NAA in this case). The USB output is directly attached to that CPU. The other CPU handles all the network activity, OS activity, HQP part, etc. In a way, I get the best of both worlds and the net result is amazing. 

 

Going back to the question about Roon. Roon does a lot of things that harm the sound quality. It performs constant network activity while you are playing, it has constant disk I/O activity, it does some processing that swings the CPU utilization, which causes noise that is audible in a resolving system. And the list goes on... Depending on whether you use a two box (a server and a streamer) solution or a single box solution like me it has different impact. But in any case, it would not be my choice for critical listening. Having said that Roon/Qobuz is the best tool for discovering new music for me. I so use it for that. 

 

P.S. A note to myself - learn what a short reply means! 

 

Wonderful response, thank you for taking the time and effort writing this.  Short reply's have a place, but a well thought out response goes a long way to clear up any confusion and is always welcome in my book.

 

Balance is key - there are what seems like unlimited settings to play with on every device in the chain and I understand this discussion isn't really geared toward the standard user "plug n play / good enough". 

 

At some point one has must find that balance and has to live within the "limitations" of their gear even after optimizing/tweaking and enjoy the system.  IOW, finding out the best source (one box optimized server, two box etc..) that works within your systems limitations/compromises is key - which you so nicely penned. 

 

Your system (your ears, your room...) might be an order of magnitude more sensitive to these source changes/optimizations then another persons system - again balance is key.  Lets also not forget about the co$t factor - if in fact that is a significant piece and or correlates to the end result. The cost factor is also sooo very relative per individual.  

 

Not knowing much about the TAS product, to your knowledge, will the TAS application only be available for taiko products?

 

[Foggie]

As a point of interest - one thing I forgot to mention above is I recently revamped my entire network - new everything (L2+L3 switching+routing, AP's) all of which was solely for segmenting network into VLANs, isolating IoT, AP's, performance, FW etc...  It was pretty intense exercise. 

 

After I completed that task and getting back to listening to my main rig, I noticed a change almost immediately and it never dawned on me all the changes I made to the network.  I was sitting there thinking, hmm something is different, the sound has changed.  Then the light-bulb went on and just realized or at least considered the fact that "duh" I just changed a whole crap load of network gear and design!! 

 

Setting up and designing this network "version", I don't think I ever once thought about or considered my listening room in the process.  Not sure why, but was hyper focused on learning and completing it since, like many, you have kids/school/work basically everything depending on the network.  Once you cut-over (after hrs on weekend) its pretty hard to go back.

[Foggie]

11 minutes ago, MarcelNL said:

@Foggie

interesting, I recently stuck my new switched in my ISP router 'isolating' audio from all else and found it nice improvement.

Can you elaborate on how you dissected your network? 

 

 

Nice.

 

Well that is a can 'o worms really😀  Basically, I went from a "flat" scheme (everything on the same network) where all traffic/broadcast is flooding the switch.  To then creating a vlan scheme to work with my enviro.  I think I created 5 vlans all with their own DHCP range/server (routing). 

1. wlan secure (AP1)

2. wlan secure (AP2)

3. wlan guest (AP2)

4. wlan insecure IoT (AP2)

5. LAN secure (home wkst/laptops etc..)

6. LAN insecure IoT

 

These vlans are between different switches as well (upstairs, downstairs etc.) in which you need to setup trunk ports to bring these together (I use fiber) and specify the valns within these manged switches.  All this then needs to come together on L3/routing and setup all the FW rules so these sep networks get access / see only what I specify.

 

I'm oversimplifying this and by no means a network guru.  I spent a considerable time creating a pretty intricate diagram for the design/layout as documenting all this is absolutely necessary.  I was also (through a friend's contact) able to provide my diagram/concept past a network guru and he gave me some feedback etc.

 

I'm not sure if that answers your question.  It would be nice to have a network section for this kind of topic as there are plenty of smart network engineers on here.  Although, I would imagine fielding network questions may get old if that is your full time job etc..

 

 

 

[Foggie]

6 minutes ago, ASRMichael said:

Exactly what I done last year. Isolating one side (home network) to other side (audio side) via fiber works well. I have my network layout in my profile on pdf. 

That's awesome.  Well thought out to say the least. 

 

11 rail PS, wow, didn't know he made that kind of PS.

 

Your H5 server build in the diagram is exactly what I am contemplating, I may hit you up with some questions!

[Foggie]

8 minutes ago, ted_b said:

I hope I am not going to hijack this wonderful thread but I've done a little searching and not found an answer.  What is TAS (Taiko Audio Server), is it both hardware and software, or a dedicated OS, or music software to replace or augment HQPlayer/Roon/etc?  I see some folks stating they are beta testing it, but also using HQplayer, etc.  A little confused, sorry for the OT.

Its a custom audio player/library application being developed by Taiko and thus far is proprietary to their platform only.

 

 

[Foggie]

11 hours ago, mrkoven said:

 

into a Holo May? ive read some folks had issues getting jcat xe to run 1.536mhz specifically with May. maybe its resolved now, not sure

I "think" it was dependent on the may FW version, not sure though