Building a DIY Music Server

[Exocer]

On 1/6/2023 at 10:54 AM, JackJohnson316 said:

I have a question about buffering and playing streamed files from RAM. If I buffer my music (streamed or local files) to ram in Euphony Stylus V4, do I need to invest in an OCXO network switch? The reason I ask is my previous streamer did not allow buffering to ram, and the network switch made a noticeable difference. I tried many different high end switches and they all sounded different. But now that my current streamer offers buffer to ram, I personally cannot hear a difference in network switches, so I sold my Etherregen and NA Eno network switch. I currently use a Jcat netcard XE with a Pink Faun LAN Isolator connected to a basic model Buffalo network switch, but the Etherregen and NA Eno didn’t make a difference IMO. Does the buffering to ram negate the need for a high end network switch? Thanks for helping me to understand.

Probably not the answer you want hear but even with buffering, power supply quality and clocking of your network switch will still have an impact on the end result.

[BCRich]

2 hours ago, Exocer said:

Probably not the answer you want hear but even with buffering, power supply quality and clocking of your network switch will still have an impact on the end result.

@JackJohnson316 @Exocer I just replaced the Hi Fi Tuning Gold Fuse in my AfterDark Hi Current LPS with a Yellow QSA; I can swear the Noise Floor got lower and the dynamics improved. Said Power Supply is powering the FireWire 800 Drive that hosts all my Music Files. Yes, I am buffering to ram.

[Nick90]

50 minutes ago, BCRich said:

@JackJohnson316 @Exocer I just replaced the Hi Fi Tuning Gold Fuse in my AfterDark Hi Current LPS with a Yellow QSA; I can swear the Noise Floor got lower and the dynamics improved. Said Power Supply is powering the FireWire 800 Drive that hosts all my Music Files. Yes, I am buffering to ram.

I’m using a Farad Super3 to power two switches and replacing the fuse for a SR Purple was a pleasant upgrade. Also changing the AC powercord is quite noticeable.

[BCRich]

1 minute ago, Nick90 said:

I’m using a Farad Super3 to power two switches and replacing the fuse for a SR Purple was a pleasant upgrade. Also changing the AC powercord is quite noticeable.

Nice, I’m lacking in the Power Cord area on this particular supply due to placement logistics. Will get that resolved at some point.

[MarcelNL]

for those in for some experimenting I'd like to suggest putting  Sprague Vitamin Q on the menu for bypass..I'm currently using a 4uF 200Vdc version on the first cap of the ULPS and having some great results, stupidly good !

[MarcelNL]

and by all means DO bypass the bypass....especially on that first C, think exotics, the rarer and crazier the better it seems...

[Gavin1977]

Is this the right CPU mount for LGA1700 on HDPlex?

 

I’m missing the M3 thumb screws, but can source these if this is the correct way to mount the heatsink. Would appreciate it if someone could share an image of their LGA1700 install if they have it…

[bit01]

11 hours ago, Gavin1977 said:

Is this the right CPU mount for LGA1700 on HDPlex?

 

I’m missing the M3 thumb screws, but can source these if this is the correct way to mount the heatsink. Would appreciate it if someone could share an image of their LGA1700 install if they have it…

I believe I followed this guide ( I bought the H5 a while back). Please have a look at the pages 6 to 9. If you are missing any parts I would ask Larry first. Yours looks like Gen 2?

 

https://hdplex.com/pub/media/image/Product/H5/H5.Installation.Manual.V3.0.pdf

 

 

 

[bit01]

@Gavin1977 if yours is the H5 V2 -it seems to have the long screw thread on the other side of the board and into the plastic rack (opposite to what you show in pic?) - i.e uses that and not the thumb screws of V3 (page 7)?

 

https://hdplex.com/pub/media/image/Product/H5/H5.Installation.Manual.V2.0.pdf

 

ATB,

b.

[Gavin1977]

Thanks - yes you have the right mount which is the H12 from the manual.  I’ve contacted Larry at HDPlex to order one.  Thanks for the feedback.

[Gavin1977]

Ok, whilst waiting for the LGA1700 mount, I thought I'd try the Taiko ATX I purchased in my 8086k rig in direct replacement of the HDPlex 800w DC-DC.

 

But it will not boot...

 

I connect everything the same, but using fis Audio ATX/EPS cables internally now.  Taiko ATX is being fed from the same Nenon V3, using the same power cable that fed the HDPlex (6-pin power input into both DC converts is the same configuration, so power supply is not the issue). 

 

When powered, the motherboard lights up but does not boot, but no leds lit on the Taiko ATX.  Turn off the Nenon v3 at the wall and then the motherboard has two red lights, and the motherboard quickly cycles through some boot codes before, one of the red lights goes off, and then eventually the other red led goes off once power from the caps in the V3 are exhausted.

 

Everything is put together correctly, I did try removing one of the EPS cables (rather than having two attached), but no difference.

 

Any ideas?

[rico25]

Same thing happened to me, with same material (Nenon V3 + Taiko ATX)

I press the "on" button of my server and nothing happened

I have to wait for 5mn and finally the server boot on its own

It is mysterious but it works :)

[Gavin1977]

oh, wait 5 mins.... well that's a new one.  I'll try

[Gavin1977]

Tried over the last 15mins…. Nice LEDs on the motherboard, but no boot sequence.

 

BIOS LED 1 is on though (located under PCIe slot)

 

 

[Gavin1977]

Apparently bios led 1 being lit is fine, just indicating which bios I have selected 

[rico25]

Arf, sorry it doesn't help , with my material it works ...

Again with my material, I notice that when everything is hot, things are easier (it works properly)

 

 

[Gavin1977]

This is bonkers behaviour… looked over my shoulder and it has switch itself on without me doing anything!

 

Got to resolve the error code now - no idea why, this motherboard is sound & working 100%
 

[Nenon]

@Gavin1977 @rico25 - what CPU are you using and what is the output voltage of the ULPS? 

[rico25]

Hi Nenon,

Doesn't change from ryzen 7 3700X, the ULPS without load is 32V :)

[Gavin1977]

43 minutes ago, Nenon said:

@Gavin1977 @rico25 - what CPU are you using and what is the output voltage of the ULPS? 

Yeah just checked - 33v out unloaded, v3 power supply seems to be working fine.

 

CPU is 8086k.  Next step is to start removing connections, or take back just to MB and build back up from there.  I’ve built a few PCs in my time, but this behaviour is odd.  Anyway error is now consistently A2, so should be possible to overcome.

 

No idea why this single component swap has caused such issue - everything else remains unchanged…

[bit01]

5 hours ago, Gavin1977 said:

Ok, whilst waiting for the LGA1700 mount, I thought I'd try the Taiko ATX I purchased in my 8086k rig in direct replacement of the HDPlex 800w DC-DC.

 

But it will not boot...

 

I connect everything the same, but using fis Audio ATX/EPS cables internally now.  Taiko ATX is being fed from the same Nenon V3, using the same power cable that fed the HDPlex (6-pin power input into both DC converts is the same configuration, so power supply is not the issue). 

 

When powered, the motherboard lights up but does not boot, but no leds lit on the Taiko ATX.  Turn off the Nenon v3 at the wall and then the motherboard has two red lights, and the motherboard quickly cycles through some boot codes before, one of the red lights goes off, and then eventually the other red led goes off once power from the caps in the V3 are exhausted.

 

Everything is put together correctly, I did try removing one of the EPS cables (rather than having two attached), but no difference.

 

Any ideas?

ATX/EPS cables were changed too?

[Exocer]

5 minutes ago, bit01 said:

ATX/EPS cables were changed too?

Good point. @Gavin1977 were you able to replicate this issue with your old cables?

[Superdad]

On 12/6/2022 at 4:48 PM, Superdad said:

Just like SFP modules with RJ45 Ethernet ports, those DACs (Direct Attach Cables) are terrible in comparison to a fiber optic SFP (or to an "AOC", Active Optical Cable, which is an actual fiber optic cable with SFP optical transceivers at hard-attached at both end).

 

Any SFP (Small Form Pluggable) that has a copper/CAT/RJ45 connection (hardwired in the case of these "DAC" variants) contains the PHY/receiver chip and a clock inside the module. The compromises are numerous, and include the crap clock, crap regulators (which must all get supplied and step down from the 3.3V that the switch/endpoint is supplying to the SFP cage), and then the extra PHY that has to go from copper to SGMII/fiber (2 differential pairs, to feed that port on the switch chip) right there in the SFP module.

So yes, the data bits will get there, but the results (owing to the host of various ground-plane-noise/clock-threshold jitter factors) won't be sounding good. 

 

Best to stick with either real fiber optic or straight RJ45 copper ports.

 

On 12/6/2022 at 11:08 PM, Superdad said:

Sorry, but despite the moniker of “passive”—versus the longer distance “active” versions—there is nothing passive about those Direct Attach Cable transceivers that stick into an SFP+ or QSFP cage.  They contain—at minimum—an RGMII to SGMII PHY transceiver chip, a 50-cent clock, a set of magnetics, and cheap power supply voltage regulators. All crammed onto a tiny board and all powered by the same 3.3V feed to the cage. So while power consumption of whatever cable they chose may be low, the noise and embedded low-frequency phase-noise/jitter performance is not going to be great.

 

By comparison, an optical SFP/SFP+/QSFP transceiver is a much more “passive” device in that its fiber to dual or quad differential pair transceiver chip runs free of any clock—direct into whatever silicon is receiving it, where it then is clocked.
[I state it this way because the output of a fiber optic SFP transceiver is not inherently Ethernet or any particular protocol. Sure it generally is fed to an Ethernet switch chip/MAC processor’s SGMII/QSGMII port, but those lovely differential pairs could just as easily be carrying data bound for some other protocol. (USB3.0 SuperSpeed being an example as it two differential pairs—one for each direction—matching nicely with SFP+ transceivers.)]

 

Major mea culpa alert! Send me eggs and I will apply to my face. 

 

What I wrote was correct about RJ45 SFP modules containing an Ethernet PHY, tiny cheap clock, and cheap voltage regs. (Avoid those!)

 

But what I said about Direct Attach Cables (DAC) was 100% wrong.  For some reason (ignorance and mind being elsewhere that day and not paying attention to words on the screen) I assumed that these DA Cables used 4 pair and were signaling Ethernet protocol--just as the RJ45 SFPs do.

Now I see (and many of the deeper tech details of it were quickly explained to me by @JohnSwenson when I ask him about it) that Direct Attach Cables use shielded twinaxial cable--just sending LVDS (low-voltage differential signaling) in fiber-mode format (a format close to SGMII, sometimes called 1000Base-X or Fibermode).  [The PCB traces at the back of all SFP/SFP+ cages in Ethernet equipment are these differential pairs going direct to an Ethernet switch chip, transceiver, or other protocol converter.]

 

Thus use of a Direct Attach Cable does--just like optical SFP/SFP+ transceivers--skip the need for an internal copper>fiber-mode PHY (as in the RJ45 SFPs).  And yes, unlike an optical SFP transceiver, there is no laser/optical>LVDS converter circuit inside--a DAC really is just twin-axial wire connected to basically blank SFP PCB. (That is why they are so inexpensive; I saw some for just $10) [This is true for the Passive versions of DAC, the Active versions of DAC--for runs longer than 7 meters--have LVDS "booster" chips to compensate for the attenuation.]

 

So indeed these DAC cables reduce the power requirements of the equipment they are connecting. How significant that is to performance and ground-plane noise is unclear. The quality of the wire used is important to keep BEC (bit error rate) low. High BEC will cause extra error-correction activity in the chips, something that runs counter to the goal of maintaining low ground-plane noise in gear. One can see that the wire gauge of a DAC gets progressively heavier--even just as the length goes from 1 meter to the 7 meter max for a Passive DAC. LVDS is great, but we are also taking about very high signaling rates, so attenuation must stay very low at high frequency and shielding should be good.

 

All that said--and my apologies again for getting it wrong the first go-round:

One should keep in mind that unlike a fiber-optic SFP/SFP+ connection, these Direct Attach Cables do not offer any galvanic isolation between the components.  So they will not interrupt the path of leakage currents or other "noise" that may be present from the upstream gear.

 

[To add that isolation back, one could design a box with high-speed digital isolators, followed by ultra-low-jitter "reclocking" flip-flops, fed by a low-phase noise clock (1.25GHz for Gigabit Ethernet), all supplied with great power networks. But that's basically a form of our EtherREGEN! (Gen2 as that's the one that will be putting Gigabit SGMII across the active-differential isolation "moat"; the first generation was just 100Mbps across the moat.)]

 

Still, a passive Direct Attach Cable may have some interesting applications and I intend to try a 7 meter DAC between two EtherREGENs in my own music system to hear the result. I presently use single-mode Finisar SFPs and fiber between them.

 

Cheers,

--Alex C.

 

[Gavin1977]

32 minutes ago, Exocer said:

Good point. @Gavin1977 were you able to replicate this issue with your old cables?

Yes - tried the old cables, same issue.

[Gavin1977]

20 minutes ago, Superdad said:

 I intend to try a 7 meter DAC between two EtherREGENs in my own music system to hear the result. I presently use single-mode Finisar SFPs and fiber between them.

 

Interesting stuff - looking forward to your findings.