Building a DIY Music Server

[Soul Analogue]

Hello from Hong Kong, 

My first post in this forum. 

I would like to chip in and share my attempt in building a PC music server, having stayed and tuned with my trusty SDtrans384 SDcard player for over two years.

I got my hands on an AMD EPYC cpu and a supermicro motherboard H11ssl-i

What I like about this motherboard is it has minimal peripherals that draws power and interfere my music playing function, all pcie and m2 slots are directly linked to CPU, no PCH!

Building a custom chassis for it...CPU to be passively cooled by a CNC aluminum block to attach to the size fins of the chassis.

and a linear power supply for all the ATX rails (3.3v, 5v, 5v standby and 12v cpu)

As you can see this project is still under development, i am not sure if it will work beautifully and sound good that surpass my beloved SDplayer

Questions and ideas are welcome!

[Soul Analogue]

Yes... every part is a challenge...

the linear power supply, the atx power up sequence, sourcing of components....

for the cooling, i dont think the cpu would be constantly full loaded, as i will be using Euphony player

and I will be running the cpu at the base frequency... no turbo and no smt (hyper threading) ... briefly checked the power consumption of the cpu running Prime95... i think the power and heat dissapation are manageable...

it is a epyc rome 7282...16c ... 2.8ghz base / 3.2ghz turbo

[Soul Analogue]

Some updates on my epyc project

Chassis and Solidworks sketch of the cpu heatsink for CNC machining 

Careful measurements of dimensions, holes and clearance are necessary

[Soul Analogue]

Details at the back of chassis

[Soul Analogue]

Please refer to the picture below:

Blue wire is -12v, which is not necessary for the computer nowadays.

Grey wire should be supplied with 5v once all the rails (3.3v, 5v, 5VSB, 12v, 12v CPU) are stable, which informs the motherboard to initialize the POST process

Green wire, you can try to just float it (ignore it) or if it does not power up, you can tie it to the Ground

 

Good luck!

[Soul Analogue]

1 hour ago, Tatomek7 said:

At the moment I just installed:

·        2 x 4114 CPUs with B14

 

Do you have an idea, if 12V 10A LPSU ( one rail ) is enough to power these 2CPUs or I should be focus on two rails, as this mobo has 3 12V “inputs”.

you should focus on two separated rails of 12v for the cpu

the 4114 has a TDP of 85w... while this figure doesnt really tell you the actual power drawn, it is your baseline reference of the current takes up by each cpu... (the actual power consumption during cpu full load is usually higher than this figure) depends on the number of cores, turbo boost, HT enabled?

by ohms law, 85w /12v = 7a ... and you shall account for the efficiency of the psu, so i would say 10a is just marginally sufficient for a single 4114.

[Soul Analogue]

12 minutes ago, dminches said:

 

You may want to just start with Windows 10 Enterprise LTSC which is already pared down to start.  From there you can turn off other services which are not needed.

if going for the windows route, i would recommend windows server 2019. The desktop experience mode is easily better than any win10 without serious optimizations

[Soul Analogue]

3 minutes ago, Mr Morris said:

Thanks both. Will get back on it. I am finding that Wiring this connector is the most challenging  part of this whole process.  Unless you have multi coloured wires.

 

I hope to post some pictures soon. 

Use colored heatshrinks to label the wires

[Soul Analogue]

7 hours ago, Nenon said:

I share mostly success stories, but there is a lot going on behind the scenes with my experiments. Here is a behind the scenes failure story... Tried to replace the clock on one of my Buffalo switches, and the switch board is fried.

 

 

The problem is I don't know why. I had someone else remove the clock, so that might be part of the problem. Also USPS was quite brutal with this package as it can be seen by the bended port. 

Well, one less precious Buffalo in this world. That specific buffalo breed is nearly extinct!

I will sacrifice another one, and hopefully it would work out next time. 

What is the symptom of it being “fried”?

no boot up with all lights on? Or no even any light when given power?

[Soul Analogue]

53 minutes ago, sig8 said:

A question for experts:

 

My current Gigabyte Z490 Master motherboard has two available 8-pin CPU power connectors. I have only connected to one connector. Should I connect to both? I have never had or seen two 8-pins connectors near CPU before. Thanks.

Most likely the two 8p are connected

you can use a dmm to check their continuity

if they are connected, there is no rational reason to connect both 8p .. just to increase the wire counts to supply the 12v....

[Soul Analogue]

On 7/25/2020 at 5:35 PM, Soul Analogue said:

Some updates on my epyc project

Chassis and Solidworks sketch of the cpu heatsink for CNC machining 

Careful measurements of dimensions, holes and clearance are necessary

 

 The CNC machined heatsink for EPYC has finally arrived...

the size fits nicely...however I just do not have a hex screwdriver long enough to drive the 4 screws...need to get one before I can properly install it and test the thermal performance...🤨

[Soul Analogue]

I have my 4kg alu block installed today

The block is bolted to the side fins of the chassis so the weight doesnt fully applied to the cpu (epyc 7282 16c smt and boost off ... All cores run at 2800mhz)

Temperatures:

Ambient 26c

Booted into windows 32c

Full load cpu by prime 10mins 42c

Full load cpu by prime 35mins 50c

Not bad for all passive😅

 

 

[Soul Analogue]

You need the wide temperature range spec from Apacer... the chips are different than the normal version

[Soul Analogue]

1 hour ago, jean-michel6 said:

Buffalo bs gs 2016 . There are many report of the sq impact of using pf ultra ocxo clock on this switch .

Did anyone tried the same modification with their less expensive ocxo clock ?

I have used Connor winfield ocxo with great result...just note that the buffalo only works with LVCMOS output clocks....CMOS output does not work

[Soul Analogue]

Continue working on the full linear atx PSU for the EPYC server. All the rails are choke filtered and regulated by my discrete voltage regulators

 

In this picture there are 4 rails (3.3, 5, 5v standby and 12v)...there will be a few more smaller current rails for add-on cards...

 

[Soul Analogue]

23 hours ago, tgb said:

Thanks Exocer & SoulAnalogue for feedback vs clocks.

 

The question is : what's important => 

- the current to power the clock ? 

Something special in the PinkFaun power supply ? far far better than a MPaudio module (LT3045 based) ?

 

The connor winfield ocxo clock draws 600-800ma at startup, while 200-300ma after 30s of start

[Soul Analogue]

Managed to take sometime to do the wirings for the LPS during the weedend.....

[Soul Analogue]

1 minute ago, Exocer said:

Looks very well built! What will this be replacing? Interested in your before and after impressions.

nth is replaced...it is just my first attempt of LPS to power the EPYC....

not bother to listen CAS with SMPS ATX psu....not even with ATX DC converter module....

[Soul Analogue]

Morning 

 

Successfully boot up the epyc server with my full linear atx psu

The voltages are pretty stable during prime95 stress test

here are some making of photos 

 

its about time to build the rails for add on cards! Stay tuned😁

cheers

[Soul Analogue]

so the main LPS unit for the EPYC server has been completed

It comprises of regulated voltages:

1) 12vdc 8a

2) 5vdc 8a

3) 3.3vdc 8a

4) 5vdc 3a standby

 

as well as 4 unregulated voltages to power the regulators inside the server chassis for add-on cards

 

Around 5 - 6 sets of my discrete regulators will be installed inside the server chassis:

[Soul Analogue]

6 hours ago, basillus said:

Hi, which PI-filter chokes are You using?

Good question...

there is no commercially off the shelf choke of this kind of specification available in the market

i have to buy separated parts (EI laminations, mounting, bobbin, etc) myself. Get the bobbin wound by local transformer maker... assemble, tune, test (for noise, both electronically and physically) each individual choke for the target voltages and current... and finally lacquer, before putting into the circuit

Pi filter chokes are technically much less demanding to make... than choke input filter chokes...

[Soul Analogue]

The hardware and PSU part of my EPYC server are almost done

have successfully separated power supply an intel i350 lan card (removing the switching regulators on the card) by 1.0v and 1.8v

the 12v will power a PCIE SSD ... probably an optane 😁 

there will be spare 5v and 3.3v rails....probably for clock modifications or usb/i2s card

[Soul Analogue]

On 10/30/2020 at 7:23 AM, Soul Analogue said:

Good question...

there is no commercially off the shelf choke of this kind of specification available in the market

i have to buy separated parts (EI laminations, mounting, bobbin, etc) myself. Get the bobbin wound by local transformer maker... assemble, tune, test (for noise, both electronically and physically) each individual choke for the target voltages and current... and finally lacquer, before putting into the circuit

Pi filter chokes are technically much less demanding to make... than choke input filter chokes...

Process of chokes stacking

[Soul Analogue]

7 minutes ago, cool_chris said:

 

This looks fantastic. 

Congratulations 👏 

Could you please be so kind to share more about how you did 1 v and 1.8v  ?

We had similar issues whith powering the Buffalo 2016 switch .

It has 1.5v and 1v sps regulators on board .

Some were able to power it up but I am not sure what is the best way of bypassing those noisy regulators.  

I have already Pink Faun clocks installed, 

but wanted to power with LPSs only bypasding this regulators

 

 

Hello Chris

I think i already post in the buffalo thread about my trial on BS2016/A

it was my finding about the current requirement of the three rails on the BS2016

I have my discrete regulator designed to go down to 1.0v and had made it work to power up the buffalo (in the trial)

I will wrap up the buffalo project after this EPYC

 

[Soul Analogue]

6 hours ago, Exocer said:

@jean-michel6 thanks for sharing your experience going the SoC route. I will most likely follow suit as I contemplate my upgrade path from an ITX motherboard an i7 8700T.

 

In other news, I have completed my MPAudio dual output LPSU. It uses a pair of ALS-HPULN modules, Mundorf Silver/Gold 15.5 AWG wire and GX-16 connectors. Output is measure at a perfect 5V.

 

I chose not to shorten the transformer wires in the event that I want to replace it later and re-adjust the module for higher voltage.

 

The intended usecase is for whichever audiophile-specific USB/Ethernet card I go with in the future.

 

This is my first attempt at this so please do not hold back on criticisms 🙂.

 

I also have a Sean Jacobs DC3 12v LPSU in the works. After all of the recent talk of 10v I wish I would have had Sean tune the modules for 10v output ;).

 

The plan, to break in both LPSUs is to try various combinations of 5v/12v on my pair of Buffalo switches.

 

 

Not bad for a first attempt.

 

What is the current spec of each rail?

 

The Primary (Live and Neutral) and Secondary transfo wires and output wires had better be twisted for "natural" RFI/EMI rejection.

 

The chassis had better connected to the mains ground, for two reasons: shielding and safety (it keeps you alive if the mains happened to be shorted and you happened to touch it)