Building a DIY Music Server

[MarcelNL]

Not trying to derail this thread, I'm new and think this is the right place for this Question.

 

Does anyone see merits in re-using an old server, there are tons out there for very little money, typically featuring single or dual Xeon 5650 or similar, 32-72Gb RAM , I was thinking to play around with one, rip out the fans tinker with passive cooling in a new chassis, add a nice linear PSU (rather a few), stick in a PCIe I2S output card, a separately powered SSD and tweak whatever OS of choice.

 

I recently switched briefly back to W10 and did not like what I heard at all (loads of tweaking to do I guess), the recent beta Daphile sounds a lot better, also briefly tried Euphony but preferred Daphile and am now in the process of making Audiolinux work to audtition that.

 

 

My current system is a NUC, running Daphile but trying to get Audiolinux to work..., fed by LPS going via I2S into my Metrum Amethyst R2R DAC feeding a Klangfilm KLV204a into my front loaded horn with open baffle using three fulrange 13" and JBL 2402 alnico tweeters. 

[MarcelNL]

That sounds very plausible and so far my findings concur with this (except the dual Xeon, never heard those...wich probably is good for my wallet).

To it appears that adding a great LPS to a low powered computer is easier, and cheaper than supplying hundreds of clean (peak) watts to a beast of a server.

Perhaps I should try pair a vintage tube PSU (Klangfilm) with a computer :-)

 

The best digital sound I have ever heard was short before the Jcat USB card died, I'm now trying to get the Pink Faun I2S card to work but it seems that it is incompatible with my (Intel) NUC (wish they had a compatibility list...).

 

[MarcelNL]

3 minutes ago, lmitche said:

 

I have a vague memory that the Pink Faun I2S card has a dependency on an AMD processor.

That is indeed what Piero told me, and he worked with PF on the OS for their server...so the search for a computer I2S continues and the PF is likely going back.

[MarcelNL]

yikes, that feels like 'if you can't beat them join them' ;-)

 

you are right, yet my fanless NUC and LPS likely are not compatible with an AMD MB...am looking into that avenue too!
 

[MarcelNL]

9 hours ago, jean-michel6 said:

I have a semi powerful server board supermicro x10 sdv-f with an 8core Xeon .

It is powered by a Sean Jacob psu .

Between the power box and server I was using a heavy gauge Neotech (awg 14) 50cm cable .

I did switch to mundorf 1.5 mm cable .

 

It did made a huge increase in sound quality .

In my opinion this illustrates how sensitive is power supply for our system and how only a 50cm long piece of wire can make or break the system .

Just out of curiosity, did you use that size for all wires and are those single leads? (meaning; do you have more leads running parallel for the same voltage)?

 

Reading up on Xeon CPU's et al I noticed that they apparently can draw hefty peak currents and afaik the current capacity for 1.5 mm (diameter or square mm?) is something like 12Amps steady state (of course with higher peak currents).

 

Anyone try a fast cap closer to the end user, as in at the power connector on the MB or even ON the MB itself? I have a hunch that adding nice caps as local buffer helps SQ. (Fleabay has some very nice Russian exotics on offer lately).  (another reason to ditch the NUC, ZERO space to work with...)

[MarcelNL]

IMO it's always good to have a strategy when tinkering, it's too easy to tinker with too many variables at a time and getting lost.

 

I try to isolate a couple of key parameters at a time, focus on getting to the bottom of each and then tweaking them before embarking on another bout of tinkering.

 

 

[MarcelNL]

3 minutes ago, jean-michel6 said:

I just used single wire . 
The tdp for this cpu is 45w .... therefore I do not think it will ever draw 12A of current . Typically I measured 1 to 3 A .

makes sense, I was not aware of the CPU having that low TDP! Ever since my Le Monstre Amp build (decades ago) my approach to anything PSU is 'think big' ;-)

[MarcelNL]

I just bought a used Ryzen 5 1600 (6 core) CPU on a GigaByte B450 DS3H and copious amounts of fast RAM (64Gb, enough for a considerable RAMDisc) that I plan to test drive in a passively cooled case, either HDPlex or Monsterlabo using a likely external LPS as my PSU builds tend to end up being LARGE. It will at least allow me to compare against a NUC Pentium, and it should be compatible with the PF I2S bridge!

 

 

[MarcelNL]

so, finally I have Audiolinux going, slightly counterintuitive setup afaik....support was great though!

 

Comparing the same setup, both at 0db reduction, swapping an SSD, Audiolinux vs Daphile makes me think that Daphile (beta RT kernel) is a step ahead in SQ.

I have not tinkered with audiolinux, no ramroot, hqplayer settings to as close to plain PCM as I can as I want to have my R2R shine.

 

AL sounds like an 80ies CD player, dynamic but flat as a pancake and no overtones, decay, original acoustics. Daphile is like listening to a very good Turntable, involving rythmical, organic, large sound stage, refined highs

 

Will do the same comparison later on the AMD board using the I2S output, am I missing anything essential with AL or are there others with similar observations?

 

 

[MarcelNL]

Listening to AL is done by HQPlayer, as I'm probably to stupid to work efficiently with Linux ; setting up things takes me forever and so far I only got the desktop version of HQplayer to work, not the embedded version. HQplayer is controlled using an app on my phone. 

 

Daphile is an all in one package, using LMS. Also not intuitive to use but IMO far easier to work with than AL (for me, I'm not passing any judgement here).

 

I have to add that switching to the latest beta RT kernel in Daphile was quite a jump in SQ, I had to switch as something changed that make Qobuz files skip, that is fixed in the beta...together with something else that changed I dare say.

 

Even the Daphile production RT kernel sounds more organic than what I so far have heard from AL, but the tinkering only just started.  Daphile sounded even better when the Jcat USB still worked, I'm hoping for another step using the PF I2S output w OCXO, but am a bit hesitant about the AMD setup...new linear PSU's needed, not an easy comparison with the NUC on a linear PSU so I probably end up comparing AL and Daphile without all the bells and wistles added except for the Linear PSU for the PF I2S and an external PSU for the SSD.

 

I tried to make the comparison as valid as possible, one key difference is the SSD, though I auditioned the new one with Daphile prior to installing AL and could not detect a major difference between the two. (the Daphile OS SSD also contains most of the music files and is EMI protected, the AL SSD also contained the music but is not yet EMI protected)

And as always, this is in my system, my ears, YMMV...

 

I thought about being biased, Daphile is free AL plus HQplayer runs at approx 300e, after a couple of swaps I'm sure it's not bias what I'm hearing.

[MarcelNL]

Halleluja, I almost needed a younger cousin to get this beast going Daphile....I'm getting old, or at least rusty at making computers do what i want

Finally Daphile is installed, now for the rest.....tomorrow, for a lack of jolt cola

 

 

[MarcelNL]

I agree that a NUC cannot hold it's water against a proper sized MB and case for the very reason everything you add sits on top, hangs out or at least needs an umbilical unless you get a larger case to put the NUC MB and peripherals in!

 

I just completed the raw POC build, installing the PF I2S card in the AMD 5 1600, which has 64Gb fast RAM on a GigByte B450 ds3H V1 MB, using the regular 'silent' PSU. Only the I2S card is fed by a 5V linear PSU, the sound is better than the NUC, just not (yet) as organic. The I2S card connects to the DAC with some cheapass premade ethernetcable, as I was too impatient to build a longer proper one.

 

I simply whacked in the Daphile SSDafter getting tired of fighting boot systems (UEFI vs legacy) etc, and it worked, the I2S card was auto detected and we have a lift off.

 

That is, all electrons need to still break in, OCXO is stone cold needs to break in many hours etc. etc. yet I do not hear 'digitis' or nasty stuff. 

 

From what I hear right now there should be plenty of potential for improvement, right now with only 15 min of playing temperature monitoring shows I run the risk of having to heat the CPU as it runs at 28'C so going passively cooled is def possible.

 

Sound quality? Lows are lower highs more transparent, there is more dynamic range, across the bandwidth, staging etc is hard to say as I am playing with only one concept speaker (room constraints in a temporary house make it impossible to add another speaker, 2.10 meter by 2.10 meter, 1 meter deep....WAF you know).

 

So now up to see if putting all in RAM brings something, or that I buy an Optane to play with (though AMD does not support it like Intel-duh- it might sound better due to a different working mechanism, and I want to heat that)

 

To the investment: I bought the computer used, some 300 euro all in all. PF I2S card 1000euro (VAT is a great invention). The real cost will come with building several linear PSU's and a passive case. I don't count the hours, this is a hobby I get enjoyment out of it and I only bill my clients. 

 

[MarcelNL]

28 minutes ago, bobfa said:

I want to challenge the thought that we can build cheaper than Commercial in many cases.   We have to know when to stop, or not.  I was looking at this from the business side. It is fun, but not viable for me.  I write all of this to explain my story and present a cautionary tale.  Have fun, know your limits.  Fail faster!  Do not run to a moving target.

What if that moving target is part of the journey towards sublime sound?

 

I like that path, as much as the sound it created... the pitfall is that no sub-project is ever finished, so there your warning is justified IMO. 

 

My target keeps moving as I learn new options for making things sound better, more natural. My audio buddy started making his own resistors and capacitor, you can do all sorts of crazy (far out too) stuff and find out what works, and that opens up new avenues.

[MarcelNL]

BTW: The POC build is 'drying up' quite nicely already, I'm getting a very tight low bass extension of a new magnitude and more organic sound. Gues that OCXO is warming up more than the CPU .

[MarcelNL]

Why go USB? If your DAC has I2S input capability the SQ improvement over uSB is an easy win at similar price level.

[MarcelNL]

anyone tried XMP on/off on their DRAM? I think it is too early to tell as my OCXO is burning in (nicely), all it supposedly does is running the DRAM at it's rated speed with tighter latency...I'm hesitant to start tinkering with all the DRAM parameters found in BIOS myself.. 

[MarcelNL]

anyone with a source for good SATA connectors? Want to make some power and data cables but can only find the crappy ready made ones.

 

[MarcelNL]

You'll need to store the data somewhere, I don't envision going the server client route if I can help it...I hear what you are saying, I listened to a couple of SATA cables and PSUs and was amazed how much that affects SQ

 

Storing all content in RAM is a bit hard, streaming means investing in more hardware as I also listened to the effects of network cables routers etc...choices...

 

[MarcelNL]

I plan to play with the M2 slot, is there any type of particular interest? would be nice to be able to power that externally....

[MarcelNL]

other than optane of course, but that is cost prohibitve for anything else than the OS

 

have one on order.

[MarcelNL]

18 minutes ago, ASRMichael said:

If really want to go for local storage, Why not go for a PCIe? If have a spare port? Like Taiko? 

That is a thought I'm entertaining...already was ogling a 1Tb PCIe SSD...

[MarcelNL]

I thought the Optane supposedly sounds better due to lower noise than SSD running the OS, I already have 64GB RAM installed and Daphile fits easily in a tiny portion of it as it is.

 

 

[MarcelNL]

Well installing the Optane was fast enough...recognized in BIOS, set to boot prio, started Daphile, low and behold I was able to copy the OS onto the Optane disk.

 

Booting happens from Optane, only snag is that my SSD appears empty (it had all music on it) and that I seem unable (for now) to copy all music files onto the SSD from scratch.

 

 

[MarcelNL]

haha, so that is where I am, a HDD on USB...

 

I found the audio files on the SSD, they are in one of the user subfolders. 

 

I will compare after I listened to the Optane...tonight when the grid is cleaner and the house is quieter...

[MarcelNL]

35 minutes ago, deama said:

I'm not sure what you mean, I've set my buffer size to 16 samples and I get a 70ms delay when measuring, but in latencymon I get similiar results to you. Latencymon I think only measures the latency between the usb and computer, which is super fast, even when using usb 2.0; the problem seems to be that windows does some processing or something and that adds lots of delay.

From what I see you are measuring ALL of the period between creating an impulse and the microphpne registering it, if I'm not mistaken there is little you can do once the impulse leaves your computer. So I assume most folks here are focussing on latency within the computer, everything thereafter is a composite of other gear and a different phenomenon.