Building a DIY Music Server

[Soul Analogue]

1 hour ago, One and a half said:

Here's a low noise design, but it's only a guide to illustrate the purpose, the voltages and currents are incorrect for ATX use. 30,000uF is substantial for power delivery, must say, not much would get through that.

 

When the load is connected across +VA and -VA, the voltage is 26V (not zero). +13V is measured at the + and PGND1 and -13V also at PGND1.

PGND1 is connected further back at the transformer to earth (ground) PE conductor (AC 12V C.T.) 

Depending on cable lengths, say 300mm or more, the load side PGND1 needs a separate wire back to the transformer earth point.

To ensure the noise cancellation works effectively, the + and - voltages need to be trimmed so they are they the same. I would expect 1mV or less would be achievable especially with lead sensing. AC transformers can balance to within 2% and work effectively at these parameters.

 

so for the computer environment, the ground of the PC will be connected to the -6V, -2.5v and -1.65v? ... I am not sure if multiple rails can be connected this way ...

there are many ways to deal with ripple noise in DC... this balanced supply theory looks like more complicated than necessary just to deal with noise...and the output impedance would be added up by the use of multiple transistors 

[Soul Analogue]

4 hours ago, Csaba said:

I would avoid purchasing from Amazon, as it is directly not available in my country. 

 

 

Check for Optane 800P...I think they are still available

I got the 118gb for OS

 

[Soul Analogue]

BTW....which is the best sounding music player for Windows? 

There are a few I can name....HQplayer, jriver, roon.... but none of them is my preference...

still exploring .....

 

[Soul Analogue]

over the weekend, I have finished up the hardware for this fanless EPYC server with full LPS

 

So the overall configuration is as follows:

Supermicro H11SSL-i

AMD EPYC 7282 16/32c 2.8/3.2ghz

Apacer Wide Temp ECC RDIMM 2666mhz 8gb x 2

Optane 800p 118gb (For OS)

Optane P4800x 750gb (For music files) (Separated 12v LPS rail)

Supermicro i350 ethernet card dual RJ45 (Separated 1.0 and 1.8v power rails)

To supply power separately to pcie cards... the golden fingers for power supply need to be masked carefully (front and back sides)

[Soul Analogue]

1 hour ago, Nenon said:

 

Very cool! And very brave of you to mod the Optane card. I hope it worked from the first time. Most people spend less on their entire music server than the cost of that card.

Power supply mod is relatively easy

What i will do in the next build is clock mod to the Optane, which is a lot more technically challenging, as the works area is very tiny, so ...

very good eyes and steady hands are required... I do not recommend non-skillful DIYer to attempt it 

(Picture with the XO, caps and resistors removed)

Made a trial... it worked after several attempts

[Soul Analogue]

1 hour ago, Exocer said:

@Soul Analogue - A job well done. In admiration of your creativity and technical skills. Do you have access to any of high end commercial offerings to perform comparisons? Also, interested in learning about what you're upgrading from with this build.

 

Cheers!

-Rob

I will send this build to my friend to do OS optimizations...afterwards i will try to arrange comparison with other highend servers...most likely SGM-Ex is my target

 

To upgrade hardware of this build... being full choke input LPS ... more separated rails and shunt voltage regulators for some areas

 

[Soul Analogue]

22 minutes ago, Nenon said:

I was not planning to spend a lot on switches either, and look at me now with not just one but two PF Buffalo switches :)). Addictive stuff... 

Same here... my next project will be Buffalo switch with full linear PSU and linear regulators

should be ready in two weeks =)

[Soul Analogue]

Last weekend, I brought my EPYC server to my friend's place for audition

It was used as server, while Memory Player output coaxial by Lynx AES as endpoint

Had some problem logging in with Roon on ipad, which was my friend's preferred player, until full update of Winserver 2019 OS and reinstall of Roon server

The sound, without OS optimization, was really promising. Very good soundstage, solid musical notes, satisfying tone and vocal...the sound outcome had exceeded my initial expectation....had a feeling of forgetting Hifi and just indulge in the music 

This fanless EPYC and Full LPS approach is definitely a way to go!

[Soul Analogue]

After the success of the first ATX LPS powering a fanless EPYC server, (refer Building a DIY Music Server - Page 43 - Music Servers - Audiophile Style)

I started to build an improved version during Xmas.

Bigger chassis, more rails, better components...aims to raise the bar in terms of PSU hardware

ATX rails:

12v 8a
5v 8a
3.3v 8a
5v sby 3a

 

Previously implemented rails:

1v 3a for i350 ethernet card
1.8v 1a for i350 ethernet card

12v 4a for optane P4800x

 

Additional rails:
12v 4a for ddr4
3.3v 1a *3 for ocxo

3.3v 1a *2 for PF i2s bridge

[Soul Analogue]

52 minutes ago, MarcelNL said:

looks great, though I would advise against stacking transformers and you may want to consider rotating some so the magnetic fields do not overlap as much as in this configuration.

Certainly i wont stack two magnetics together... 

Spacing will definitely be given

[Soul Analogue]

Here are SOME of the discrete voltage regulators for my new EPYC server

[Soul Analogue]

18 hours ago, seeteeyou said:

those of us who are willing to go for WS C621E SAGE + dual Xeon Silver could get a copy of TAS at some point

 

Could you elaborate on this?

[Soul Analogue]

been working slowly due to other projects....
Now almost complete the 13 rails Linear psu for the improved EPYC build

A few eye candy shots to share...

 

the 14 pin plug is a hell to work with... need to be very careful with the pin assignment 😅

and making the umbrilical cords takes some good efforts

[Soul Analogue]

So my improved AMD EPYC build has almost been completed..

 

This is a shot of discrete regulators inside the main chassis

And a shot of the regulators under testing

[Soul Analogue]

I see so many of you are talking about choke input (LC) filter for taiko ATX module... but is there anyone really put into practice or a simple experiment?

The taiko ATX can easily draw over 10A current at 30v...

The choke for LC filter has to be purposely designed and substantially made (huge size) as it is dealing with all the AC ripple from the rectifier....

I am 100% sure that the 10A spec Hammond choke (just a mere 2kg) will not serve this purpose, or the choke will be as noisy as a chatterbox even under a 2A load

In my experience, a choke built for this LC purpose needs to be at least 7-8kg weight to be reasonably quiet and have a meaningful ripple reduction capability...

 

 

[Soul Analogue]

21 minutes ago, Dev said:

how about this monster at 72lbs ? 🤪

 

https://www.mouser.com/ProductDetail/Hammond-Manufacturing/195R20?qs=7rwefYdqV%2FZHBBiCJpS93w%3D%3D

 

 

 

May be ok.... the size looks more practical 

but as I said , the choke must be purposely built for LC application..I do not see it is mentioned in the Hammond datasheet 

 

[Soul Analogue]

19 minutes ago, MarcelNL said:

I was also wondering how a 5mH choke would do anything serious, my tube amp has a custom 400H (@100mA) HiB core choke weighing a ton , of course that PSU uses far less capacitance than this PSU due to the mercury filled tube rectifiers but still.

 

5mH choke for 10A current is not bad already

[Soul Analogue]

14 minutes ago, RickyV said:

Okay this not a tube amplifier but a server it does not hum. What would the continuous current draw of the server be? I don’t think it’s 5 or 10 A continuous. That’s where I think the big capacitors come in, to deliver the spiky current draw of a server.
On a side note the Taiko uses a Lundahl choke not sure what it’s current rating is.

 

Yes...the server does not draw that much current constantly...but I think nobody would accept a PSU / hifi equipment rattles intermittently (during necessary loading times)

When designing a PSU, always consider the max load, otherwise reliability issue will arise

 

SGMS Extreme is only using capacitor input filter on the PSU.... so the requirement / spec on the choke is much relaxed

[Soul Analogue]

4 minutes ago, RickyV said:

choke on the right

 

 

 

Capacitors on the left ...which is right after the Recifier diodes = Capacitor Input filter....

Clearly NOT choke input filter that I mentioned above

 

[Soul Analogue]

2 minutes ago, RickyV said:

Okay so what is that Lundhal thing doing? Found a better picture.

 

 

 

 

It is a Capacitor Input CLC filter topology...the L does help on ripple reduction than Capacitor only filter

[Soul Analogue]

1 minute ago, Dev said:

What is the circular thing beside the Mundrof ? That seems like a choke as well ?

 

 

 

I think it is a film capacitor to bypass the 47000uf electrolytic

Definitely not a choke

[Soul Analogue]

1 minute ago, RickyV said:

?? Is the L in the CLC not a choke?

In this threat we are discussing CLCLC filters, taiko dc-atx power supply for nenon’s diy server.

YES the L is the Choke

But having a choke does not mean it is a Choke Input filter, which is a choke Right after Rectification, No capacitor in between

CLC filter is still a Capacitor input filter.....

Please differentiate these topologies.... 

[Soul Analogue]

23 hours ago, Tatomek7 said:

 

Hi @Nenon, Do you know what is power requirement to boot C621 MB? Hammond choke is rated with 5A, do you know, what current is taken from PSU by Taiko ATX board? In the meantime ( waiting for final information ) I'm considering to base rectifier on diodes, do you know what amperage will be needed?

 

I saw Tatomek7 quote on Nenon's post dated 16 May 

 

"The second version (aka v2) consists of: IEC inlet --> Fuse --> Soft start --> Transformer --> Mosfet rectifier --> Hammond choke --> 3 x Mundorf HC 22,000 uF (each bypassed with VCAP ODAM 0.1 uF film cap) --> Output connector. This is an improvement over v1 in my opinion."

 

this is a choke input (LC) filter 

 

and this thread developed with Hammond Chokes to be used in this kind of topology

Thats why I was asking if anyone had tried that

[Soul Analogue]

23 minutes ago, Dev said:

 

Ok, you have a point. I have been thinking CLC filter in my head all along and that is how I am planning to implement it with the Taiko DC-atx. I will let Nenon clarify. But even with CLC filter, the question still remains weather a 10mH/5A choke be sufficient ?

It depends on what you mean "sufficient"....even if the power drawn by the computer is higher than 5A...the choke core saturates and gets a hot ..... the ripple current cannot be controlled ..... but the PSU will still work if the voltage still within the operation voltage of the DC-ATX...

if you are going to power the full CAS machine, I would definitely suggest at least a 10A choke...

the lower DCR of higher current choke will give lower PSU impedance and hence better dynamics

[Soul Analogue]

Guys..looking at the photos, I reckon that the Hammond 159zj is way too small for the purpose....

and its only rated at 5A... whats the max target current it is running??

the choke like that will seriously impact the dynamics, and hence the life of music...

 

Try a bigger size choke...at least 10A spec...I am sure there would be a breath-taking improvement 

 

This is the unregulated LPS I build for my friend FYR