Building a DIY Music Server

[Nenon]

17 hours ago, internethandle said:

@Nenon have you ever attempted something like this motherboard memory power bypass that was discussed some time ago in the Irish Tir Na Hifi forum? They were using LT3045 regs back then and PHD supplies etc. to directly power the RAM sticks. Something I haven't seen done before or since.

 

Here's the thread:

 

http://www.tirnahifi.org/forum/viewtopic.php?t=3267&start=90

 

Some pics from the above page:

 

 

 

4 hours ago, Ben-M said:

Not that the method is the same, but separately and directly powering the ram was a top flight modder's mod back 4-5 years ago on the JPlay forums. Everyone who did it said it was a critical mod, but I never saw it catch on with any "commercial" music PC offerings and it seems to have drifted out of the limelight now. 

 

I always thought it was an interesting sounding mod. 

 

This is interesting. I thought about it at some point but never tried it. It would obviously require a careful study of the motherboard design to figure out how to cut the ATX power and how to feed the RAM with a LPS without impacting the sensor / protection triggers. 

 

This would be quite difficult / costly to do on the Asus Sage motherboard I am using now. It has a quite complicated design, and I have 12 DIMMs. Already destroyed one CPU by accidently shorting two memory pins. With a $650 motherboard price and $550 per CPU street price, a few mistakes can make this experiment quite costly. I am not willing to do that. But it's probably worth trying on a cheaper and less complicated motherboard.

 

At this stage, I think that full linear power regulation is not necessarily the best thing in the world. While a linear power supply is the most important component in the digital source, I find a mixture of linear regulators and switching regulators to sound better.

Perhaps that has nothing to do with linear vs. switching regulators but more with low powered vs. high powered processing. High power processing requires... as you can guess - more power. And it is practically impossible to provide the required Amps of low voltage (i.e. 1V let's say) fully linear regulated.

 

Imagine having to provision 100W at 1 Volt of linear regulated power to the CPU? That would be 100A of current. And let's say you are using regulators that need to drop 5V. That would be 500W of heat dissipation for your linear regulators. Passively cooling 500W of heat dissipation? Not sure what it takes to do that, but in my mind I imagine a heatsink the size of a speaker :). The good news is RAM does not require that many watts.

 

I am yet to hear a fully linear regulated computer or streamer that sounds better than some of our DIY attempts with high powered CPUs that use quite a few switching regulators. There are companies like Sonore, SOTM, etc. that try to do that, but those are in my opinion far away from what we are achieving with our DIY experiments here. But again, that's not comparing apples to apples. In fact every time I have replaced a switching mode power supply with a good linear regulated LPS, I have heard an improvement. So that experiment is definitely worth a try. 

 

Please post here if you try this. It would be nice to see a step by step guide and what the outcome was in sound quality. There are quite a few people here that can do it. 

 

[Nenon]

27 minutes ago, mikicasellas said:

I would like to improve the quality of the power that is supplied to my server ( CPU, MOBO, JCAT XE USB, ER ), in my mind there are multiple permutations and im trying to find out where could be the sweet spot: where to put each LPS and IF it could be benefit from another one in the mix without bleeding too much $.

 

These are my actual LPSUs:

 

HDPLEX 200W

PH SR4T 19V

DXP-1A5DSC Ultra Low Noise

DXP-1A5DSC Ultra Low Noise

 

 

 My server is built as follows right now:

 

1- AMD 7 3700X processor ( powered directly from HDPLEX 200W )

2- Asus Rog Strix X470-I Mobo ( powered from HDPLEX 200W thru a HDPLEX 800 DCATX converter )
3- 8GB Apacer Industrial RAM

4- Octane M10 32 GB

5- HDPLEX H5 Fanless case

6- Ghent JSSG360 internal and external cables

7- JCAT USB XE card ( 9V from Paul Hynes SRT4-19V and then DXP 7V input and 5V output )

8- UPTONE EtherRegen ( 12V from HDPLEX and then DXP 10.5V input and 9V output )

 

 

A ) - Powering JCAT XE USB, CPU, MOBO and ER with my actual PSUs:

 

SCENARIO #1A:

HDPLEX 200W powering ER & JCAT with DXPs

HDPLEX 200W powering CPU directly 

PH SR4T "try" to power MOBO thru HDPLEX 800W DCATX

 

SCENARIO #2A:

HDPLEX 200W powering JCAT with DXP

HDPLEX 200W powering CPU directly & MOBO thru HDPLEX 800W DCATX

PH SR4T powering ER with DXPs

 

 

B ) - Powering JCAT XE USB, CPU, MOBO and ER with adding a PSU (SJ, PH, KECES to mention some):

 

SCENARIO #1B:

HDPLEX 200W or PH SR4T powering JCAT with DXP

HDPLEX 200W or PH SR4T powering ER with DXPs

HDPLEX 200W powering MOBO thru HDPLEX 800W DCATX

SEAN JACOBS, PAUL HYNES or else PSU with a 12V 5A output for the CPU

 

SCENARIO #2B:

PH SR4T powering JCAT with DXP

HDPLEX 200W powering MOBO thru HDPLEX 800W DCATX

DUAL SEAN JACOBS, PAUL HYNES or else PSU with a 12V 5A output for the CPU & ER with DXPs

 

This is so far what it comes to mind, some of you have experienced various of these combinations

 

Suggestions are welcome where to best start... it does not have to be all at once but i can start building up as my money allows until server is at its best.  I appreciate your feedback

 

Thanks and keep safe !!

 

 

 

I personally don't find mixing many different power supplies to give the best results. You are mixing multiple grounds from different power supplies, and that is not necessarily a good thing. The ground plane in a power supply plays a big role. You can build two power supplies from the same schematics, for example, one with neat star grounding and one where the ground goes all over the place, and typically the one with the better grounding technique would sound better. You can imagine what happens when you mix 2, 3, or even 4 different power supplies, each with its own ground.

 

Have you considered selling all of these and getting a single multi rail LPS? PHD and SJ can do that. It would be more expensive but that would be my recommendation. 

[Nenon]

3 hours ago, Gavin1977 said:

@Nenon For your dual CPU Asus c621 build, how is it's HQPlayer performance? Can you run Sinc-M, or Sinc-L at DSD512, with some of the more demanding modulators like ASDM7EC?

 

I don't know as I don't upsample. Those CPUs don't have super high clock speed and may not be the best for heavy Hqplayer usage.

My guess is if you push them too hard, you would run into some cooling problems with the HDPlex heatsink... And you would need a high current LPS. 

[Nenon]

On 7/14/2020 at 11:10 AM, mikicasellas said:

 

If i keep only HDPLEX & PH SR4T by now, am i going into this type of issue ?

 

Do the DXPs contribute to this problem ?

 

Thanks

Sorry, I missed your question. I would not call this an "issue". But to me it's not ideal to mix multiple power supplies that all connect to the same ground eventually (i.e. the motherboard ground). 

My comment was more generic that I find the ground to be quite important and prefer not to mix the grounds from multiple power supplies. That's given all other factors are equal. If other factors are not equal, like in your case, I can't tell what's better. You would have to try for yourself. 

 

In terms of grounding, your 2A scenario looks better.

On 7/13/2020 at 4:38 PM, Nenon said:

SCENARIO #2A:

HDPLEX 200W powering JCAT with DXP

HDPLEX 200W powering CPU directly & MOBO thru HDPLEX 800W DCATX

PH SR4T powering ER with DXPs

 

 

But I also have a feeling that your 1A scenario is worth a try. In this case, you are mixing the grounds, but it could be that the SR4T is performing better on the DCATX. I don't know, because I have never compared the SR4T to the HDPlex 200W directly. 

On 7/13/2020 at 4:38 PM, Nenon said:

SCENARIO #1A:

HDPLEX 200W powering ER & JCAT with DXPs

HDPLEX 200W powering CPU directly 

PH SR4T "try" to power MOBO thru HDPLEX 800W DCATX

 

When you power the CPU directly, the ATX connector typically does not need a lot of power. It needs around 30W in my case. 

It could be that the SR4T is a better LPS and that brings more than the multiple grounds take away. The important part is the final result, and only you can test and decide what sounds better in your system.

 

 

[Nenon]

Welcome to the forum @Soul Analogue.

This is a very enthusiastic project. Looking forward to hear more about it. I don't think we have tried an EPYC CPU. Which model did you pick? I guess the passive cooling might be  a challenge. 

[Nenon]

7 hours ago, 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

 

Looking good. Keep updating us. Thank you for sharing. 

[Nenon]

6 hours ago, Mr Morris said:

Quick question. I am wiring up my 24 pin ATX connector back to my regulator boards. What happens to the grey and the green?  Thanks 

 

6 hours ago, Mr Morris said:

. .. and the blue!

 

I am assuming you mean pin 8 (grey), pin 16 (green), and pin 14 (blue)?

 

 

Pin 8 connect to the +5V DC rail.

Pin 16 controls the ATX PS on/off. You don't need to connect it. 

Pin 14 is not needed. Same for PIN 20 - not needed. 

 

[Nenon]

7 hours ago, Tatomek7 said:

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”.

 

5 hours ago, Soul Analogue said:

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

 

Keep in mind that this motherboard uses only one 12V rail. The 12V from the ATX connector, the 2 x 12 Volts on the EPS connector and the 12V on the third connector are all connected in parallel. If you use two 12V power supplies, there might be some strange consequences as you are basically putting the output of two power supplies in parallel. 

 

This was one of the strange consequences in my case :):

On 4/22/2020 at 5:42 PM, Nenon said:

 

If it helps, I ended up using a one rail 12.5A LPS that can deliver up to 45A of instantaneous peaks. I feed the 12V on the 24-pin ATX and the two EPS connectors from that LPS. It's similar to the one @elan120 built but with DC4 specs and a few other extras. Normally, it draws around 6-7A while playing music. But that depends on your software too. 

 

[Nenon]

4 minutes ago, Exocer said:

 

Interesting. 

 

I have a two rail LPS on the way. One 19v rail will feed an HDPlex DC-ATX power supply and the other rail, 12v, will go directly to CPU.

 

I am now wondering if this will be cause for concern with the ASUS Strix Z390-i you selected for the first project in this thread...

 

I suppose I could repurpose the 12v rail for other devices if I must.

Let me clarify...

My post was about powering two CPUs on the same board. Those CPUs don't have isolated power rails, and you can't use two dedicated rails (i.e. one per CPU). Hence, I recommend using one EPS rail for both CPUs. It just happened that the 12V EPS rail on this particular motherboard is also shared with the 12V on the 24-pin ATX connector.

 

You are good with your two rail LPS for the ASUS Strix Z390-i. That's what I would do in your case too. My post was about a different case. 

[Nenon]

3 hours ago, exeric2 said:

Hi, newbie here that has been following along. I plan to reproduce Nenon's Asus Crosshair version with Ryzen CPU. I have many of the parts in hand and others are on order. One thing I haven't been able to order is the HDPlex 800 watt dc-atx converter. They are not in stock anywhere and no update from them when they will be in stock. I know Nenon was not impressed with the sound using the HDPlex 400 dc-atx but I'm thinking I may try it.

 

Here's the question: I'd like to use the 400 dc-atx but not use +12V output from it to the motherboard. I have a dual rail LPS coming with 12V and 19V rails. I'd originally planned to use the 12 rail to only go to the EPS connector. I'd like to now feed the 12V atx connection from that same LPS 12 rail. Would there be a problem doing that. The only one I can imagine happening is with different grounds going to the atx connection on the motherboard, one from the LPS 12 rail and one from the atx ground that is being fed from the other 19v rail of the same LPS. Could this be a problem? I realize this might be going into the weeds but does anyone know what some of the different grounds on the atx connector are linked to. If there is one specifically linked to the 12V input on the motherboard atx connector I could supply that ground straight from the 12v LPS rail...

 

The Asus Crosshair does NOT share the same 12V rail between the EPS connector and the ATX connector. I would recommend you power up the ATX connector from the HDplex 400/800 and power the EPS connector from your 12V rail. There is a benefit to isolating the 12V CPU/EPS rail from the rest. 

 

All motherboards I have seen have a common ground plane between all the rails. If you really want to try to feed the 12V ATX from the same rail as the EPS, you can of course try that. But I don't recommend it. 

 

As for trying to get a HDPlex 800W DC-ATX, email Larry. He gets small batches and sells them via email. Have him add you to the waiting list, and you would be able to get one faster this way. 

[Nenon]

9 minutes ago, exeric2 said:

Who is this infamous Larry? Does he have an email address? It seems everyone here knows about him but me.

 

He is the man running HDPlex. 

https://hdplex.com/contact/

 

[Nenon]

1 hour ago, jean-michel6 said:

however there is no Sean Jacobs psu above 5A

 

Check with him. He has one but does not advertise it. This thread might be of interest to you if you have missed it:

 

[Nenon]

7 hours ago, jean-michel6 said:

I have one question which puzzled me . On the beginning of this thread you are detailing the construction of a server that you built using three dc3 reg to supply the dc atx for this board . 
I am surprised to see that there is no boost modules for these dc3reg . 
The dc3reg by itself is capable of only 1.5A power which is quite low you need the boost module to go to 5A .

Sean had two types of DC3 regulators at the time - 1.5A and a more powerful one. I called the more powerful one "5A" in my posts. 

 

On 1/10/2020 at 11:15 AM, Nenon said:

There will be 3 regulator boards in this chassis, all of them feeding a 20-pin ATX connector on the motherboard. They are: 3.3V (1.5A), 5V (5.0A), and 12V (1.5A).

 

I think those don't necessarily provide continuous 5A, hence Sean is offering the boost module for 5A applications. But the 5V ATX rail does not need continuous 5A. It only needs to be able to handle occasional peaks over 1.5A, so that arrangement worked fine. I had tested it on other builds and knew that worked fine.

 

My friend already had a SJ 12V LPS with a boost module, which he used for the EPS:

On 1/10/2020 at 12:01 AM, Nenon said:

The EPS connector will be powered by another 12V Sean Jacobs DC3 LPS he already has.

 

[Nenon]

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. 

[Nenon]

Thank you all for your concerns and willingness to help. This is a normal day in the life of a DIY-er :). We just don't post too much about failures and only share success stories. I decided to switch things around for a moment. Give me a few days to get one working, and I will post pictures and instructions. I'll then try to inspect closer what happened to this one. 

[Nenon]

I just replaced the clock on another Buffalo switch and it worked like a charm. More to follow. 

[Nenon]

1 hour ago, sig8 said:

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.

 

The main reason some motherboards have more connectors is to be able to provide more current to the CPU, for more power hungry CPUs or for overclocking. The typical wire going to the EPS is 18 AWG. 4 x 18 AWG wires give you about 12 AWG combined gauge. But if you use both connectors, that would be 8 x 18 AWG wires, which is around 9 AWG.

The motherboard Taiko Extreme has 3 connectors - 2 x 8 pins, and 1 x 6 pin, plus 2 of the wires from the ATX connector go to the same +12V DC rail. So that's total of 13 x 18 AWG wires, which is around 7 AWG combined wire gauge. No wonder why Taiko Extreme users report that thick power cords work really good!

 

In your case, you have to look at multiple factors. How many Watts or Amps is your power supply? How thick are the power supply wires? What processor are you using? How are you using the CPU? Are you overclocking or underclocking? Are you upsampling? And so on...

For most cases just using one EPS connector is good enough. I suspect that's the case for you as well but don't have enough details to tell for sure. 

[Nenon]

5 hours ago, Nenon said:

I just replaced the clock on another Buffalo switch and it worked like a charm. More to follow. 

5 hours later, and I am still listening to Qobuz :). That's a good sign. That switch with a PinkFaun ultraOCXO clock is sounding pretty good. I'll post more after I spend some time with it. But I am planning to replace the clock on one more Buffalo; and try two Buffalos with upgraded clocks (one is for a friend, but I will have the opportunity to test it). Those clocks are expensive, but they don't fail to surprise me every time I try them. We are reaching the price of the M12 Gold switch.

As a matter of fact, if the timing works in my favor, I would be able to test the M12 Gold switch in my system while I have the two Buffalos with upgraded clocks. I am super curious what happens when you connect two Buffalo switches with ultraOCXO clocks, one M12 Gold, and the etherREGEN together :). It's complete madness, but why not if I have a chance to do it... It would also be interesting to compare the M12 Gold with those Buffalos with upgraded clocks. 

 

However, now that I am playing in the bigger league, I may not have the time (or the motivation) to test the smaller 8-port Buffalos.

 

On 7/7/2020 at 4:17 PM, Nenon said:

Buffalo switches - does the 8-port model sound like the 16-port model? Time to find out.

 

 

But first, some serious burn-in time needed.

If anyone is interested in those two 8-port switches, send me a PM. Too many Buffalos in my backyard :). 

[Nenon]

Here are more details on my modified Buffalo switch. 

 

1. PinkFaun ultraOCXO clock - 25 MHz - https://www.pinkfaun.com/shop/clock/67-2916-ultra-ocxo.html

It was difficult to remove the stock clock, but once I did the rest was easy. A couple people recommended the ChipQuik kit. That stuff is great and it works really good on chips with many legs that are exposed outside the chip. But because the four legs on this clock are under the clock, you don't really have access to them. 

The easiest way I have found so far was to cut the clock in half with sharp clippers and desolder each half individually by heating the two legs with my soldering iron. 

Here is a close up of the pins I used for the PinkFaun clock:

The filtering capacitors and resistors are not required. You can leave them or remove them per your preferences. 

I used a very fine 24 AWG solid silver wire in enamel, similar to what AudioNote sells. 

The clock is mounted on special vibration damping standoffs that decouple the clock from the switch board. 

The bolt with a red circle around is using one of the holes on the switch board. I drilled holes in the switch board for the other 3 holes. If you are careful with the alignment, you will end up with 3 safe to drill places. Just make sure you use non-conductive nuts or add non-conductive washers. I added plastic washers on the bottom in my case. Otherwise the nut would short one part of the PCB to ground. I believe this is why my previous attempt failed... I think the board I thought that failed might be working, but the mounting nuts caused a problem. 

 

2. The clock requires a dedicated 5V power supply. The quality of this power supply is crucial. I used a GX16 connector and 18 AWG Mundorf silver/gold wire.

 

3. I made a small PCB board to handle some extra capacitors. It mounts on the standoffs from the removed switching mode power supply. That should close the gap between the Buffalo and Melco switch. This board was done for my DIY hobby experiment only, and I won't talk much about it. 15.5 AWG Mundorf silver/gold wire is used on both sides of this board to connect it to the DC connector (GX16) and to the switch board. 

 

4. I added two additional filtering capacitors. One is a 2.2 uF Mundorf silver/gold/oil film cap. This is a decent film cap but not my favorite. There is no space for Duelund CAST caps in this chassis. The other one is a 100uF AudioNote Kaisei I had laying around. I may tweak those in the future but it's a good starting point. 

 

5. Grounding post. I used WBT Nextgen copper grounding post, just because I had a spare, and I like those and a 24 AWG stranded Neotech 7N copper wire. I don't like using solid wire for this type of grounding. I also don't like plated wires or much thicker wire. That should be a good balance. I ground the chassis and the clock, but will be experimenting more with that. For example I may ground the minus from the power supply to help shunting some of the noise. That's an area I would be exploring further.

I have a Taiko SETCHI D-3 to try and a Synergistic Research Active Ground Block SE. 

 

6. Chassis damping. This chassis is not designed for audio and wrapping with bitumen-based sound dampening material helps with the chassis vibration. 

 

7. Gaia feet. Not on the main picture, but additional vibration treatment with Gaia footers was also done.

That's it. Those are all my tweaks. 

The switch will be powered with a dual rail Sean Jacobs DC4 LPS and of course my favorite silver/gold DC cables. 

 

How does it sound? I am quite impressed so far but would like to live more time with it so it would properly burn-in and my bias settles down. But I know there will be some impatient members who would send me a PM asking about the sound, so I decided to take some notes and share my first impressions.

 

Initial impressions of the Buffalo with upgraded ultraOCXO clock.

My network during the test:

Arris SB8200 modem --(copper)--> Ubiquiti Router --(copper)--> Buffalo switch --(fiber with Fintech transceivers)--> etherRegen --(copper)--> server. 

I kept everything the same, just swapping two identical Buffalo switches, one with the stock clock and one with a PinkFaun ultraOCXO clock. Both Buffalo switches were powered up with the same 12V LPS by Sean Jacobs. The 5V for the clock was powered by a LPS-1.2 (temporarily until I get the parts needed to adjust my rails). 

The switch with an upgraded clock is much better. I can’t listen to the stock clocked Buffalos anymore . It's hard to describe the change. 

There is no particular area that was improved. I can't tell there is better midrange, or better highs, or more dynamics, or anything like that. But the overall coherence of the sound has improved. The improved timing has a big impact on how the instruments sound. The difference is similar to listening to a garage band vs. a professional band playing the same notes. The music sounds overall more enjoyable. 

What’s really cool is the level of PRAT you get from bad recordings. It does not make bad recordings sound like good recordings but it makes them so much more enjoyable. It’s like you listen to a bad recording on a good reel to reel. And that's an interesting thing as well - with this switch with a modified clock, my digital is a step further away from sounding digital. And that's for streaming Qobuz. The sound is not lean like with many digital components with improved clocks and linear regulators sound. The body of the Buffalo is still there. If anything has changed, there is maybe a level of harshness that has disappeared and the bass is more clear but still deep and impactful. 

I have found that switches also impact local storage playing for some reason. Don't really expect that changing the clock impacts locally stored files, but I think it does. But my focus is on streaming music from Qobuz. 

I still have a lot of tests to do. An interesting experiment would be what happens when I plug my server directly to the Buffalo with ultraOCXO clock. Right now the ultraOCXO clocked signal goes from the Buffalo to the etherREGEN, which has a good clock but not as good as the ultraOCXO. Does that mean I am losing some of the benefits of the good clock? Needs to be tested by connecting my server directly to the Buffalo switch. On the other hand, the etherREGEN is designed to minimise noise... However, in my system using the B-side of the etherREGEN is not as good as only utilizing the A-side, so I don't go across the moat and don't utilize that part of the etherREGEN design. What does that mean for the noise generated by these switches? Would the Buffalo generate more noise? TBD. One thing that may help is to shunt the power supply on the Buffalo to ground. That's something I am definitely going to try. 

But as we speak about differences between my modified Buffalo and the etherREGEN, we need  to keep things into prospective. The upgraded clock costs more than two etherREGENs. The power supply is another seven etherREGENs. The rest of the tweaks also cost as much as an ehterREGEN. And ever since we posted about those Buffalo switches here, the price went up 3 times, and they are still extremely difficult to find. I expect now PinkFaun will run out of clocks :)). And I need one more, so don't buy them all just yet. 

 

I will be modifying the clock on one more Buffalo and hopefully will be able to do some interesting tests.

For example, what happens when you connect two Buffalos with upgraded clocks together? How do they compare to the M12 gold switch which I have access to and will be testing soon? And what if you combine two Buffalos with ultraOCXO clocks with the M12 Gold switch? I mean other than getting divorced and filing for bankruptcy...

[Nenon]

On 8/7/2020 at 9:05 PM, Nenon said:

An interesting experiment would be what happens when I plug my server directly to the Buffalo with ultraOCXO clock. Right now the ultraOCXO clocked signal goes from the Buffalo to the etherREGEN, which has a good clock but not as good as the ultraOCXO. Does that mean I am losing some of the benefits of the good clock? Needs to be tested by connecting my server directly to the Buffalo switch. On the other hand, the etherREGEN is designed to minimise noise... However, in my system using the B-side of the etherREGEN is not as good as only utilizing the A-side, so I don't go across the moat and don't utilize that part of the etherREGEN design. What does that mean for the noise generated by these switches? Would the Buffalo generate more noise? TBD.

 

Just a quick update:

I swapped the positions of the etherREGEN and my modified Buffalo switch. Now my server is connected directly to the Buffalo like this:

Arris SB8200 modem --(copper)--> Ubiquiti Router --(copper)--> etherRegen (fiber with Fintech transceivers)--> Buffalo switch--(copper)--> server. 

 

That turned out to be a shocking improvement! I expected a small difference that makes me do many A/B comparisons until I decide if I like one better than the other. Not at all. 

 

I just picked a random Qobuz track for the test. It just happened to be this one:
artist: The Ray Brown All Stars

track: You Don't Know Me
album: Don't Forget The Blues

But I also listened to different styles music after that to make sure it's not just that track that sounded better

 

It took about 5 seconds from the track to hear how much better the new combination was. By the 14th second I was convinced what I heard the first 5 seconds was true. After 30 seconds there was no doubts at all. We've all been there, and you probably know what I am talking about. Towards the end of the track I knew I had a new reference! Obviously I went back and forth 5 or 6 times to double/triple/quadruple/quintuple check and never doubted even for a second which one was better. Never heard Qobuz sound so good! Going back to do some more critical listening now... 

[Nenon]

59 minutes ago, lwr said:

Is the EtherREGEN re-clocked?  Were both of the Buffalos re-clocked, or was it just one?  If just one Buffalo, which one in the series was the re-clocked one?

 

Both your experiments and your reports of the associated SQ benefits are a real inspiration.  Thank you!

 

Sorry, I only used one Buffalo switch and one etherREGEN (corrected my post, it was a copy and paste error). Like this:

1 hour ago, Nenon said:

Arris SB8200 modem --(copper)--> Ubiquiti Router --(copper)--> etherRegen (fiber with Fintech transceivers)--> Buffalo switch--(copper)--> server. 

 

Stock etherREGEN (no external clock) and the highly modified Buffalo I posted about. That sounded much better than 4 x Buffalos daisy chained together. It's really really good. The best network I've had in my system.

 

I am still planning to modify a second Buffalo and get the M12 Gold switch to compare. I have been avoiding the M12 Gold switch, because I was pretty sure as soon as try one I would buy one. But now that I have a new reference, it would be interesting to see how they compare. 

[Nenon]

2 hours ago, genvirt said:

Hi,

I hesitated to put this questions here or to open new thread, at last I thing it’ll be maybe helpful to others...

I decided to renew my audio PC that is 4 yo, and now I’m looking for “ right “ configuration, I’m not a PC guru, so I need an advice from experienced members.

All what I need to know for now that which Motherboard, CPU and Memory to choose, and here is I’m getting a bit lost.

Marcin from JCat once advised me this configuration :

 

- supermicro  X11SCL-F

- Xeon E-213

- Apacer DDR4 ECC UDIMM 2x4GB or (2x8) server/workstation extended temp. range RAM (Maybe this configuration. was changed already...)

 

Someone point me and advice to choose CPU with T and 35w for not being to hot and Gigabyte mini Itx and Corsair Vegeance LPX 2666mhz DDR4  

 

Here Nenon with a wonderful set up using that stuff:

- ASUS ROG Crosshair VII Hero full size ATX motherboard 

- AMD Ryzen 7 3700X CPU (65w)

- Apacer ECC RAM
 

So, where to go?
where is advantage or disadvantage of each advice?
what will sound better (I understand that all this is subjective , but...)
 

I see that almost nowhere 35w cpu is using - why? Is it no good? Wrong? Not enough ?

I’ll go to use tree cards (usb, net and I2S) that will power by external supply.

 

And question regarding ATX and EPS powering:

I noticed that recently each one being power by external supply, where is benefits if , for example, when using HDPlex 800w DC-ATX converter (with external power supply 19v?) is more than enough to power both...

 

Thanks a lot and sorry for maybe silly questions, but should to choose the better configuration...

 

 

 

There is no simple answer. There are so many factors. Imagine going to a car building forum and asking what car should you build yourself without giving any context, without knowing where you live, what your skill level is, what do you need the car for (i.e. driving to work, racing, offshore driving, etc.), without knowing what your budget is and so on. Your question here comes across in a similar fashion. 

 

I have no idea what system you have, what music you listen to, what your personal sound preferences are. Are you trying to listen to streaming services like Tidal and Qobuz? Or are you trying to listen to locally stored files? If you have a lot of local files, are they hi-res, or CD quality? Or DSD? Then we need to look at what DAC you are using and what inputs it has. The interaction between your server and your DAC is one of the most critical points. There is just a huge number a variables. 

 

If you want a server that can do almost everything perfect, you can buy a Taiko Extreme for €25K. And still that may not be the best server for heavy upsampling algorithms. But I am guessing people come here to read and discuss, because going the DIY route can potentially give you more value for the money you spend. But that means we need to do compromises. And this is where it becomes very complicated.

 

What are the right compromises for you? 

If you were building a car and live in a south carribean island, one compromise that you could make would be to not install heating system inside the car. But if you live in Alaska, you may want to do a few things to the car in addition to a heating system like heated steering wheel, heated side mirrors, etc. That's how we tailor our DIY servers to our needs. 

 

If you are doing heavy upsampling for example, you need a powerful CPU with a higher clock speed. But what determines if you need to do upsampling? Well, the DAC you are using is a good starting point. If you are using a NOS DAC most likely you can skip the upsampling part. If you are using a DAC that upsamples everything inside the DAC to high DSD format, then maybe you should consider upsampling as an option. But what would sound better? An upstreamed signal from a not so good server? Or a native format from a better server? That would depend on the DAC and the server and which one is more efficient and can deliver better quality results when the two work together in a team. 

 

The CPU you pick would determine the power supply requirements. A 35W TDP CPU server would have significantly different power requirements than a dual 85W TDP CPUs server. Power is the most important factor. A 35W TDP CPU with a really good LPS may sound a lot better than a dual CPU (85W TDP each) with a bad power supply. And vice versa. 

Since a good power supply is typically more expensive than the computer components, it's always good to start with a budget and narrow down what power supplies you can afford within your budget. 

 

Whether you upsample or not would also help you determine what digital interface card you can use. If you have to upsample to a higher sample rate format, USB is probably your best option. If you are using a NOS DAC you can consider an i2S interface or S/PDIF. Again, it depends on your DAC and how your server and DAC interact together in each case. If most of your music is DSD, that can also make your decision, and you would most likely prefer to go USB. 

 

Do you want to optimise it for streaming? If you do, you may want to get a JCAT Net Femto card for example. And it's probably a good idea to get another power rail for the JCAT NIC. That also means that you may want to look into motherboards with at least two PCIe slots (one for the network card and one for the digital interface). At that point you may also want to look at the motherboard architecture. Are the PCIe slots directly connected to the CPUs? Or are they going through a chipset? Or is it a combination?

 

Then you can look at the storage strategy.  If you optimise your server for streaming, it would have good quality network access. In that case you may consider using a NAS to store your local music. But that also depends on the software you are using. If the software is buffering the music to RAM before playing, it would be okay to use NAS. But if your software is streaming from your NAS in realtime, that typically makes NAS storage sound not so good. You can consider changing the software (in which case you have to revise everything else you have done). Or you can do some other local storage. HDDs vibrate and that does not have good effect on the components in the same chassis. We are talking about clock replacements in this thread... vibrations impact the clock quality. SSD drives don't have any moving pieces, so no vibration there, but they generate some electrical noise. You can use a dedicated LPS, but the noise gets back to the motherboard via the SATA cable. And speaking of SATA, you need to look at your motherboard architecture again and see if that's even a good option. USB drives can potentially generate noise on your USB. And if this is where your DAC connects, you may want to be careful with that. I like NVME storage and use two 2 TB NVME drives on a PCIe card with VROC RAID0. But not all CPUs support VROC. And then again you need to have an extra PCIe slot, etc. etc.

 

What OS will you be using? How does the OS interact with your hardware. And the list keeps going on and on. I might have covered about 1% of the possible things to consider when designing a server. Every CPU sounds different. Every motherboard sounds different too. Every OS sounds different. Every software player sounds different. Imaging how complex it would be to test every combination for every scenario at the same time in the same system with the same power supply, the same OS, the same player, etc. It's virtually impossible. 

 

There are many DIY server recipes on this forum. Do some reading and research. Try to understand what the person who designed the system was trying to achieve. The novel thread has so many documented experiments over several years. Pick one recipe and give it a try, get your hands dirty. As you get more experience you can go back and tweak things. DIY is constant tweaking. That would be my advice to you.

 

Both servers are pretty good (the one Marcin recommended and the one I did). You can build both of these and with all other factors as similar as they can be, the one that has better LPS would sound better. Or you can build both of these with the same power supply, and I bet I can let you pick one (any of the two), and I would be able to make it sound better than the other with just making Operating system level changes on both. I am only saying this to illustrate how complex this is and that the end result is a combination of all pieces combined together.

Here is an example - take a Taiko Extreme server, load AudioLinux with Roon with no tweaks and see how much worse it sounds than the highly tweaked custom Windows. All the magic would be gone. Extreme users even reported that all the magic was gone when Roon released an update... On the other hand, you can read in the novel thread how many people migrated their servers from Windows to AudioLinux or Euphony and how much better it sounded. It is a typical example where hardware and software need to be fully optimised to work together.

 

In the case of the Extreme server, the custom Windows is configured to take advantage of the two CPUs. Every process and every thread is controlled by affinities and rules. This essentially has a server/streamer replicated on the same server. 

 

Finally, here is a study from Emile he posted in another forum:

https://www.whatsbestforum.com/threads/taiko-audio-sgm-extreme-the-crème-de-la-crème.27433/page-163#post-639048

Quote

A visual analogy on your television set would be:
-Intel Xeon scalable 10 core as a reference, controls to neutral
-AMD Epyc +1 on the brightness control -1 on contrast and +2 on white level or on backlight intensity on a LED screen
-AMD Threadripper +2 on the brightness control -1 on contrast and +2 on white level
-Intel Xeon scalable 12 core +1 on white level
-Intel Xeon scalable 8 core, controls to neutral, but reduced staging/dynamics
-Intel i7-7700K (EVO) +1 on brightness, +1 on white level, similar staging/dynamics to the Xeon 8 core

 

But don't take those numbers in a vacuum. The motherboard has as much influence on the sound as the CPU. You can't try an AMD CPU and an Intel in the same motherboard. So it's not clear how much of the influence on the sound during these comparisons was because of the CPU and how much was because of the motherboard. I would look at these numbers as the combination of the two (CPU+motherboard), but he did not list the motherboards used. 

 

As for the RAM, there is a general consensus that the industrial grade wide temp grade ECC Apacer sounds better than the general commercial RAM. One of the reasons for that is because Apacer uses A-grade Samsung chips. 

 

[Nenon]

1 hour ago, djwalter2000 said:

I continue to read this thread with great interest. It's inspired me to try some new things and continues to refine my thinking and approach. 

 

My question is to those who are building single CPU systems. My approach has been to use an ITX form factor with onboard CPU (in my case a j1900) and DC barrel connector. I'm using streaming services exclusively and don't have a need to upsample or otherwise tax the processor with 'robust' software. 

 

Taking this path, I have an easier time with cooling, case selection, and setup overall. Powering the board with an external LPS requires a little less customization but I still power the SSD, JCAT Net and Element H USB cards with their own dedicated linear supplies, making it a 4 rail system. I'm not just after the easiest approach. While I don't have all of the technical chops many of the contributors here have, I like a challenge. But more may not be better if the CPU has a somewhat small job. In a perfect world, we'd all have Taiko Extremes. 

 

For anyone who may have taken a similar low powered CPU approach but moved on to a more robust CPU/MB build, I'd like your thoughts and experiences as to why this change provided better SQ, not just more overhead to run more complex software. 

 

If this topic has been covered extensively elsewhere, just let me know.

 

Thanks!

 

Just like I mentioned in my previous post every motherboard sounds different and every CPU sounds different. It would be quite wrong to generalize between those that have a DC connector and those that don't. Similarly to generalize the motherboard form factor. 

 

If that helps, I have tried several ASUS ROG mini-ITX motherboards and their bigger full-size ATX siblings. I prefered the full size ATX. The full size ATX also gives more PCIe slots, and you don't have to use riser cards/cables that definitely mess up the sound quality. 

 

I have also tried low powered CPUs and high powered CPUs and almost always prefered the higher power CPU. The only time I would prefer a low-powered CPU would be if I don't have an adequate LPS for the high-powered CPU and I can only power the low powered CPU. 

[Nenon]

4 hours ago, iCarus said:

It's a very interesting thread but you can get first class HiFi using two used Intel Atom boards and a Linux distro. 

 

Please share. I have a couple of them in my closet and was never able to get first class HiFi, no matter the rest of the hardware, power supplies, operating system, software, etc. Actually at the time I used them I thought they were pretty good, but after another year of experiments, I think they are good but not even remotely close to some of the builds discussed here. 

[Nenon]

On 8/26/2020 at 2:16 AM, Soul Analogue said:

 

 

Wow