Building a DIY Music Server

[lmitche]

On 2/5/2020 at 11:45 PM, Nenon said:

 

Thank you. That makes sense, because these are also the two steel enforced PCIe slots.

In that case, the plan is:

- PinkFaun Bridge in PCIEX16/X8_1

- JCAT Net Femto in PCIEX8/X4_2

- Optane in M.2_1.

- Leave M.2_2 unused.

 

Actually I need to check how M.2_1 connects. Hopefully not through the chipset. I may end up using M.2_2 instead. 

You can view the device hierarchy in windows and linux. Use the connection view in windows device manager or the lspci -t command in linux. Anything at the first level is CPU connected.

 

Otherwise you need a block diagram for the motherboard in question. They can be hard to find.

[lmitche]

On 2/5/2020 at 5:55 PM, Nenon said:

What I don't like about the case is that this plate needs to come out completely every time you want to add or remove a PCIe card. But I haven't assembled it yet, so maybe there is a workaround this. It also feels a little cheaper quality than the Streacom, and I wish the top and bottom plates were thicker. 

Having just completed a build in a Hdplex H3V3 case, the ability to remove the backplate is a unique and a much appreciated aspect of this new Hdplex case. Compared to a Streacom F9, the Hdplex was much more flexible enabling use of all the pcie slots of a MicroAtx motherboard in a smaller case than the H5. While I understand that one could consider the aluminum materials used in Streacom to be of higher quality, the Hdplex uses a mix of steel and aluminum in a practical, highly evolved design. Hdplex cases will be my first choice going forward.

[lmitche]

On 2/9/2020 at 4:52 PM, Nenon said:

As for the removing plate - I am glad you like it, but I find that part of the design the worst thing about this chassis.

Funny, in my mind the removable back plate is the best feature of this case. The Streacom has a similar challenge with it's 2, not 4, usable pcie slots with the back plate riveted in place.

 

We will need to agree to disagree on this point.👍

[lmitche]

4 hours ago, Nenon said:

The NUC sounds good but the builds in this thread are on a completely different level. 

I don't necessarily disagree with this, but I think some more detail is important. A well tuned and powered NUC can be very hard to beat.  A major reason that a NUC sounds so good is that they are based on a System-On-a Chip(SOC). NUC SOCs are designed for laptops, and are intended to be powered by batteries, and automatically switch off unused circuits to conserve power during laptop operation. An example of this is support for temperature-compensated refresh (TCR) which reduces SODIMM power utilization on some models when enabled in BIOS. Likewise the reduced distance between components within the chip reduces latency, and impedance adding to SQ.

 

Mini-ITX, Micro-ATX and ATX motherboards most often use two chips, the processor and Platform Controller Hub (PCH) that do the work of the one chip in a NUC. Some motherboards allow the disablement of the PCH and doing so greatly enhances SQ and puts these boards on, at least, an even footing with a SOC based NUC. Lastly many Intel processors for the consumer market include on board GPUs. Most AMD Ryzen and Xeon based solutions do not. The absence of a GPU is audible, and on Intel chips disablement of the GPU through software is possible as a GPU and monitor is not required for music playback. This works on NUCs as well.

 

From what I read, Emile's Taiko Audio Extreme appears to have a NUC based competitor in the Grimm audio MU1. It will be interesting to see how this competition between these two Dutch companies shakes out.

[lmitche]

2 hours ago, auricgoldfinger said:

 

How can one determine if this is possible on a particular motherboard?  Is it disabled physically?

 

 

Is the GPU disabled in the BIOS?

 

Thanks, Larry.

In some motherboards the PCH can be disabled in Bios.

 

The GPU can be disabled in the OS.

[lmitche]

3 hours ago, gererick said:

Presumably whether or not one can or cannot disable the GPU in the operating system depends on the operating system. 

 

For example, per Nenon, Euphony is a closed operating system and changes are not possible.  Euphony is the best-sounding operating system, in my setup.  Regardless of GPU enabled or disabled.  And Euphony sounds best with a powerful processor, which I am guessing rules out a lot of NUCs.

 

Then there is the question of whether (i) motherboards that allow the PCH to be disabled sound better - after the PCH has been disabled - than (ii) other motherboards that do not allow the PCH to be disabled.  The motherboards in clause (ii) may have other SQ advantages that more than offset.   

I would expect that the Euphony people can add GPU disablement as an option.

[lmitche]

1 hour ago, beautiful music said:

 

That's quite interesting things Larry.

 

Curious to know what is your Server+LPS configuration now and what is the specification of the more powerful Server that you compared it to your NUC server?

Luckily, I have several Intel and AMD server machines and many NUCs along with a truly ridiculous number and variety of linear power supplies. The number of network boxes, wifi, copper and fiber is embarrassing. All this stuff can be used in single machine or server / endpoint configurations. Roon running on a heavily modified version of Audiolinux is my daily driver.

 

My goal is not to build or own a Rolls Royce. A BWM is good enough for me.😁

 

 

[lmitche]

5 minutes ago, beautiful music said:

 

What's the best NUC did you find so far and how did you power it?

I prefer the i7 NUCs based on the NUC7i7DNB boards. The design is a couple years old now. The newer NUC lines are a bit of a mess at the moment and I have no experience with them. There are no obvious substitutes for the older product line.

 

IME a PH SR4 is one of the best LPSes for a NUC I7.

[lmitche]

25 minutes ago, tims said:

HI @lmitche - OT, but you mention laptops in yout post; any thoughts on how to improve a bog standard laptop for audio use (other than software i.e. fidelizer). 

What about replacing the laptop's switched PS with a linear supply? Will this improve anything?

This is indeed OT, so quickly, a well tuned OS and good lps will probably greatly improve the SQ of a laptop.

[lmitche]

34 minutes ago, gererick said:

There probably are not a lot of instances where Euphony has added a bios option at a customer's request.

Euphony has added many features that came from ideas first discussed in these pages.

[lmitche]

3 hours ago, OAudio said:

Having said this, research and direct experience here is limited to Intel Chipsets so its not sure if AMD PCHs (X570) behave differently.

Here is a two year old article discussing the difference between the Intel and AMD clock implementations.

 

https://www.anandtech.com/show/12678/a-timely-discovery-examining-amd-2nd-gen-ryzen-results

 

It is clear that the OS and BIOS configs have a big role in determining which clock is used. It looks like the industry is moving to TSC clocks onboard the processor chip.

 

My AMD motherboard running Audiolinux shows the TSC clock is used, which is a clock onboard the Ryzen 7 2700 processor chip, so the PCH is not used. In my case the PCH is disabled.

 

#cat /sys/devices/system/clocksource/clocksource0/current_clocksource
tsc

 

 

 

[lmitche]

1 hour ago, OAudio said:

Is there a mode that can explicitly shut down the PCH on this type of AMD system ? I'm just thinking that the PCH will still be live and there will be some traffic between it and the CPU, just not much in the case of you set up.

Yes, some AMD motherboards have a entry in the BIOS to disable the PCH.

[lmitche]

1 hour ago, taipan254 said:

 

This was a really interesting read. In particular, I was struck by the description that HPET forced on in the OS for intel based systems increased latency but reduced errors. Is it wrong for me to extrapolate that, in the context of audio processing, that forcing HPET on in the OS would result in higher latency but less variation in latency given fewer errors? This seems to be true for both AMD and Intel (see below).

 

 

If so, this would correlate with @ray-dude's findings that he presented in part two of his SGM Extreme review. Lastly, if true, this reasoning would indicate you would want HPET on. @lmitche, what am i missing here? I have an old AMD system running AL. I'll mess around tonight with my settings and see if I can spot any differences. 

 

 

 

 

You are not missing a thing. Please let us know how your experiment with clocksource goes.

 

Larry

[lmitche]

1 hour ago, MagnusH said:

If you only need ethernet for your audio needs, and use a JCat NET Femto card with external power, will you still get better sound from all the other tweaks?

Yes, most of the tweaks suggested here are additive.

[lmitche]

On 10/26/2020 at 6:56 PM, OAudio said:

 

I think it would be good to see measured performance specifications provided as well.  

 

As a general observation there are some large ticket LPSs out there for audio systems but few vendors provide measured specs for products. At the price levels we are operating at, it's not unreasonable to be looking for similar measured performance to lab class LPSs from companies like R&S Tektronics and Agilent.

 

So ball park or better than:

Sub 50 uSec 60% load transient recovery times.

Sub 2mv rms ripple and noise.

Sub 2mv rail sag from 10 to 100% of rated load.

(For 6 to 10 amp rated supplies)

 

I know from developing supplies there are many layers to a good design but my experiance is that these measurements do correlate with SQ and could help better inform people's decisions.

 

Are these targets or actual measurements?  Nonetheless this is great to see. It has always struck me as odd that power supply measurements for audio applications are not regularly quoted. While listening will always inform a purchase decision, power supplies bring out the objectivist in me. There is no excuse for not providing these details for any audio supply.

[lmitche]

On 10/29/2020 at 11:49 PM, seeteeyou said:

since a more powerful CPU should sound (much) better under most circumstances

As a community service, we should stop promoting the falsehood that more CPU power means better SQ. It does depend on circumstances, and in most cases is just untrue.

[lmitche]

2 hours ago, dminches said:

 

In trying to refute someone's opinion you have injected your own.  I don't agree that "in most cases" it doesn't sound better.  It all depends on the design.

 

In my experience, if you can properly cool and power the higher powered CPU it will sound better than the lower powered counterpart.

 

What OS and player software do you run?

[lmitche]

57 minutes ago, dminches said:

 

I am running Euphony / Stylus.

 

I have a Xeon-based server with a Sean Jacobs DC4.

 

Thanks Dminches,

 

My experience is that in a computer running Euphony/Stylus more CPU power means better SQ. I believe there are many here that have had a similar experience with Euphony such as yourself.

 

The original comment was "since a more powerful CPU should sound (much) better under most circumstances". This is a complete generalization that makes no mention of Euphony. My objection is with the generalization.

 

The server mentioned in the post doesn't run Euphony. The post states it comes "with custom BIOS / stripped down version of Windows 10 IoT Enterprise LTSC" running on a Pentium J4205. While the Pentium is far from the fastest Intel processor, to imply that SQ must be below Euphony running on the fastest Intel processor is conjecture at best. We wouldn't know unless we have a listen. That is straight logic, not an opinion.

 

I believe people should listen to a solution before sharing an opinion based on some "theory" or extrapolation of others experience. There is no substitute for "hands-on" empirical evidence. An opinion on sound quality based on anything else is a disservice to the community.

[lmitche]

On 11/4/2020 at 1:23 AM, Nenon said:

Power supplies are the most important part of a DIY music server

Hi Nenon,

 

Do you mean that power supplies are the most important contributor to SQ in a music server?

 

Larry

[lmitche]

4 hours ago, Nenon said:

 

It's the sum of all parts after all. There are countless number of tweaks one can do in a music server. 

 

On all my previous builds I prefered Linux. I am quite familiar with Linux and have no problems compiling custom kernels tweaking the OS, etc. But it was a never ending story. I wanted to spend my time better. This is when I switched to AudioLinux which made the unlimited number of options somewhat more limited. But eventually I switched to Euphony and never felt the need to go back to hardcore OS tweaking. That was on my AMD and intel i9 builds.

 

With my dual Xeon server, I prefer a stripped-down version of Windows LTSC and process Lasso to configure process affinities. This sounds better than Euphony on that particular server. Also an advantage of Windows is the ASIO driver that you can install and configure. 

 

It takes a lot of tuning to get the best SQ out of Windows or Linux. After working on Windows for years, I gave up trying to turn an operating system designed for office automation into a real-time OS for music playback. This experience makes me reluctant to follow Emile and Taiko Audio, and recently Nenon into the Windows IOT platform. Windows DRM is another gnarly issue as well.

 

Euphony is locked up, so there is no way to add value. In my experience it sounds OK, but I have done better by tuning Audiolinux. Several of my customers have tried Euphony as well, and have validated this conclusion.

 

Today dozens of changes to Audiolinux and the Roon environment, along with complimentary BIOS changes, yields a combination that sounds spectacular. The resulting OS image works equally well on both Intel NUC and AMD platforms. The past few months has brought some of the most significant contributions to SQ discovered to date, all from software configuration changes. It is hard to say where future enhancements will be found.

 

Nenon is right, this a game of the sum of the parts. Based on recent experience, I'd put software at over 50% of the contribution to SQ with hardware/cabling and power supplies at 25% each.

 

SQ = .5(SW) + .25(HW/Cabling) + .25(LPS)

 

What do others think about this formula?

 

Larry

[lmitche]

8 hours ago, gererick said:

Clarification, on the server Nenon built for me, Euphony was superior to Roon (without having to fuss with the OS). That same server, using Euphony, was materially better than a low-power audiophile arrangement with tweaked AudioLinux and Roon.  Nenon’s server had better wire and better power supplies, so there were other factors besides just software.

Hi Rick,

 

Yes, well the current Audiolinux/Roon build shipped with systems today has SQ that is light years ahead of the version delivered to you over a year ago. Much has changed since.

 

Also I believe Nenon's custom build included a custom power supply built from Sean Jacobs regulators. As you say, that may have had an impact on SQ.

 

Just to be clear, are you still using Roon?

 

[lmitche]

8 hours ago, Dev said:

In terms of tweaking, Linux is a bigger rabbit hole than Windows if you are proficient not only in terms of OS customization but also you could modify at the source code level which is not possible with Windows.

 

when I moved from Windows to Daphile, I started liking what I heard but that slowly turned into a desire of wanting to do something more myself and that lead me to Audiolinux as Daphile is closed door. With Audiolinux, you would tweak something and then listen to compare even if I was happy prior to tweaking. It’s just the desire of wanting something more which slowly turned into an obsession. This went for a while, until I realized that I was no longer enjoying music as I used to but enjoyed tweaking on almost everyday basic. Nothing wrong tweaking, but this is not what I wanted and I am glad that there is Euphony to rescue. Not only Euphony sounds better than Audiolinux in “my” system, it also takes the headache of tweaking out of the equation by completely closing its doors (well only a very limited set which is very much manageable). 
 

it’s possible that Audiolinux hasn’t been tweaked to the extent it needs to be to compete with Euphony but I have no time nor desire to do this endlessly. This is why I am also reluctant to experiment with Windows. It’s my poison so far ☠️ But who knows I might change my mind later 😉
 

 

@lmitche you haven’t mentioned what those tweaks are which makes Audiolinux better than Euphony. Maybe it’s a trade secret or something you are selling ? Is there a tweaked version of Audiolinux that can be paid/bought ? I am not looking into NUC and low power stuff - that road has met a dead-end a while back and have no desire to go back. If you are selling something packaged (s/w only) similar to Euphony, please post some details as I am interested (and so are many here I guess), otherwise it really doesn’t help much for those of us who have no desire to tweak Audiolinux.

 

 

on SQ and how much does OS, cabling and h/w, power supply helps - IMO, they are all equally important and I would put equal weights on each of them. To get the best SQ out of a particular system, they all have equal role to play and has a cumulative effect to the overall sound.

 

Hi Dev,

 

It is indeed easy to buy and install Euphony, plus the results are good.  Clearly there is a market for that, and it is terrific that it works for you and so many others. I am happy to build solutions using Euphony and listen to it, as reference, on a regular basis.

 

Yes, I do ship most systems here with custom Audiolinux almost always running Roon.

 

Anyway, it seems this is mostly a hardware thread, so I will take this discussion elsewhere. I was just curious to learn the current state of thinking about OS/Music player tuning from Nenon and others interested in the impact of software. I am detecting a trend moving away from Roon led by Takio Audio, Grimm Audio and others, so the next few years will be interesting.

 

Personally I'd like to see Roon SQ improve, as I love the user interface as a music discovery tool.

[lmitche]

11 hours ago, dminches said:

it allowed me to increase the max CPU frequency and turn on things like hyperthreading in the BIOS. Before this I had to limit things. 

Dminches,

 

FYI - Over here there is no better way to suck the life out of the music than to enable turbo mode and multi-threading.

 

Is this observation with Roon or Euphony Stylus as player? Intel or AMD?

[lmitche]

5 minutes ago, dminches said:

Euphony, Intel Xeon

 

 

Which music player on Euphony?

[lmitche]

52 minutes ago, Nenon said:

I believe the formula, if one exists, would be a lot more complex, and would depend on the hardware being used and some other variables. 

 

Last time I chatted with @lmitche he reported a great success with inexpensive Celeron-based computers and heavily tweaked software and BIOS. This is great and although not my cup of tea, there is a much much bigger market for that than some exotic cost no object builds I have been trying to do.

I think with hardware like that (i.e. referring to inexpensive Celeron-based computers), the power supply can only make so much difference, but then you can fine tune the whole system with software and BIOS settings. I can see where the 50/25/25 comes from. And it's probably about right in this use case.

 

This is also quite typical for low powered computers. I can confirm that because I had the same experience. 

 

On the other hand, we have high powered computers, and those have different impact ratio of power supply / software / hardware/ cabling IMO. The idea there is to unleash all the power capabilities of the computer. But you can only do that if you have a really good quality power supply. Enabling hyperthreading, Turbo, unrestricted CPU speed is typically a must in those servers. That brings the life back in the music and makes everything sound much more dynamic, transparent, etc. Once I heard that, there was no way of going back to a low power CPU. Also, once I heard the dual Xeon, I am not even interested to go back to a single CPU. Although there are some new developments in CPUs that look interesting. 

 

The main problem is that unleashing the processing capabilities means the server draws more power. Most linear power supplies become very noisy when you draw that much current. Some can become as noisy as (or even more than) SMPS at high current. Not to mention the heat they generate which also could have an impact on the sound. That noise is easily audible as what I call digital harshness. I am very sensitive on that digital harshness and can't stand it. When you don't use a really good power supply, you have to start restricting the CPU computing power (i.e. lower the clock speed, voltages, disable turbo, etc.). Imagine a knob that has "more dynamics" on one end and "less digital harshness" on the other end. Let me illustrate that:

(hopefully the developer of the app I stole this picture from does not mind it)

You are essentially trying to find the balance between the two to compensate for the power supply noise.

 

Things change with better power supplies. 

 

In those higher powered CPU systems, the power supply makes the biggest difference. And like everything else in audio once you get a really good power supply, the software tweaks are even more audible, because everything becomes more transparent. 

 

@dminches has an amazing analog system, and he was pretty confident digital could never sound as good as analog. He has a Lampizator Pacific DAC, which is also the reference DAC in the Taiko's system. The main thing he changed in his system was the power supply. We talked again after his new LPS burned-in, and his story changed from "analog for critical listening, digital for background music" to "enjoying both and it all depends on the recording" - quite a different story before and after his LPS. A 20A+ capable top quality power supply opens up a lot of opportunities for powerful computer hardware, which generally speaking I prefer too. 

 

I am not trying to promote a specific power supply, and I try to be very careful given my industry affiliation (which makes it difficult for me to comment on some topics). I am just trying to make a point that the software/hardware/cabling/PS ratio is different for different hardware and different systems. 

 

It's not even that simple. When I started building high current CPU music servers, going from a mediocre power supply to a really good power supply made much bigger difference than the software. But once I had a really good power supply, the software changes made much bigger difference than going from a really good power supply to an even better power supply. It's all relative. And different things make different impact depending on the current stage you are in.

 

No need to argue on this - I am sure everyone's opinion is right relative to their current stage. And no need to try to put an universal formula - if we do, it would be so complex to cover all the variables, that we would all need a math degree to understand it. At the end, it is the sum of all parts and everything needs attention. 

 

Okay, enough morning blah blah from me :). Cheers.

 

Hi Nenon,

 

Thanks for the thoughtful reply. Over here 4 to 8 core AMD servers are typically powered by two 200 to 300 watt supplies, one at 19 volts and one at 12 volts, so yesterdays comments are based on experience at this power level. I will let you characterize this as high or low. Actual processor power use is between a steady 8 or 17 watts during playback.

 

While we talk a lot about power supplies, little discussion happens about power demand. Once a track is selected, music playback is a simple monotonous process of copying, and perhaps uniformly transforming, bits from the source device to the output device. Assuming no up-sampling or filtering, there is little to do as there is constant, unchanging demand for processing power during this process. With most players, and for a given file resolution, processing demand does not change with music content.

 

Power supplies handle constant current demand much better than changing demand, and if a system is tuned in this way the need for exotic power supplies is less important.

 

If the design of a player changes processor demand with music content, then it makes sense that the supply of processing bandwidth enabled by multi-threaded and turbo mode is required and that the quality of the power supply increased.

 

But for clarity, the idea is that better managed power demand diminishes the impact of the power supply and shifts the balance to software as a larger contributor of SQ. The impact of the power supply quality is still there, it is just diminished.

 

Sadly, there used to be many people on what we called CA that would have been interested in this discussion. These days there seem to be few. I do not mean to be argumentative, and apologize if this was the wrong thread for this discussion.