Building a DIY Music Server

[OAudio]

 

Intel "client" level Optane SSDs go end of life this month.

 

Just a heads up for anyone considering using an Optane drive and who has not heard about Intel's change of market strategy on Optane. I got a distributor bulleting late last year to saying some Intel SSDs were to go end of life in Q1. In the announcement recently are all of  the "client" level 900p and 905p series Optane devices are withdrawn this month (Feb). Going forward the available high capacity, none hybrid, Optane drives are all to be "data centre" units which are significantly more expensive than the withdrawn 900p and 905p ranges.

 

It might have been mentioned in other threads I haven't seen, but thought I would call it out here is case anyone is thinking about using Optane SSDs in the near future. Right now would be the time to buy if you need one of the last 900p / 905p series drives. 

 

 OAudio  

 

     

[OAudio]

22 hours ago, seeteeyou said:

FYI - here's another Gallium Nitride (GaN) option with 500W / DC 28V output to either Taiko ATX or HDPLEX 800W DC-ATX etc.

 

https://product.tdk.com/info/en/catalog/datasheets/pfh_e.pdf

https://product.tdk.com/info/en/products/power/tec_data/ps_pfh.html

https://www.transphormusa.com/en/news/tdk_lambda_pfh500f_power_module/

 

250 bucks for the module and 500 bucks for the evaluation board respectively

 

https://power.sager.com/tdk-lambda-pfh500f-power-supply.html

https://www.digikey.com/catalog/en/partgroup/pfh500f-series/79643

https://www.digikey.com/catalog/en/partgroup/pfh05w-evaluation-test-boards/83877

They are nice looking units but perhaps not very well suited to powering a quality DC ATX. The PSU's spec's show droop rate performance under load and in particular the ripple and noise specs would be a concern. And there can also be combined noise issues with  an SMPS followed by DC ATX buck conversion.

 

The specs on the supply linked show:

Droop Rate (-xDx-R suffix) 50 mv/A
Ripple & Noise 400 mv

(Both for the 28v version).

 

They are big figures for feeding a DC ATX in a server that's focused on sound quality.

 

I spent quite some time a few years ago looking for a SMPS solution to power DC ATXs. There would be much going for such a setup if it could produce really good sound. None that I looked at worked well feeding DC ATX's for sound quality however. The problem came down to having two unlinked switching controllers, one in the SMPS and a second buck controller for each rail of the DC ATX that followed. Both of stages vary switching frequency and or duty cycle as they see fit to track the current load being consumed by the PC. The result for the combinations looked at was mess of noise both on the input to the DC ATX and on it's output. On the DC ATX outputs there was a combination of the harmonics of the switching frequencies of both stages. Never manged to get a nice sound from the combinations tried.

 

If you can, perhaps stay with regulated linear or passive filtered power. It's a well trodden path but for good reason.

 

Something with droop under load in the very low mv/A range and combined noise plus ripple of <5mv. Transient response isn't quoted for the linked PSU but if it is published for another supply a <100us recovery time for a transient of 5 amps would be a decent place to start.

 

 

 

 

 

 

[OAudio]

3 hours ago, AME said:

Looks great,

 

I am however curious which USB chipset will be used. I understood that it will not be using the ASM 3142.

 

 

Looks very much like it uses the same host controller I also alighted on when screening available chipsets, the μPD720202. 

[OAudio]

6 hours ago, Nenon said:

The power supply that I built for @dminches was the game changer for him. I had a similar power supply powering my dual CPU server before switching to the Taiko ATX. And here is the truth about linear power supplies: they work amazingly good for low current applications. But when you get to high current draws like the dual Xeon CPUs in the Extreme, regulated LPS is not the best technology to use. I would use and recommend the Taiko ATX with the unregulated LPS for the dual CPU project. In fact I would put this against any regulated LPS, no matter the price, including the high current DC4 I built for @dminches. I compared the two in my system and had a hard time deciding. But the Taiko ATX + unregulated LPS for the dual Xeon project sounded more transparent.  And that is an important part of voicing the server. 

 

 

@Nenon, All,

 

I really would not want this pass into forum folk law, without putting an alternate view and note of caution forward. I think using higher current headroom as key parameter to select a power supply is akin to using amplifier wattage alone to compare sound quality of power amplifiers. It might not be a very good idea. Here is some of what I found designing supplies for c621 servers and why I say this.  

 

Important to remember first that there isn't anything mystical about c621 boards, they are just incrementally slightly larger then previous Xeon architectures. Threads on other web sites make it sound complex but they are just computer PCBs. These boards have finite and quantifiable load requirements. Provided that a supply be it linear or buck or any other type fully meets these the current load requirements of its rails the supplies current capacity head room can be as little as 1 amp and the board will not be bothered, and sound quality will be fine. A reasonable margin of current headroom is good thing as confidence factor but I have run dual core c621s at with 10 amp rated EPS supply designs and the sound has been spell binding. What really matters is on how the supply for the board performs in other more critical areas. 

 

So what matters for sound quality ? Based on developing and testing these supplies (background given below) the following: 

 

• 1 amp headroom above maximum motherboard rail's peak transient load level - kidding... a few amps for safety is a better idea :-)
• Transient response fast enough across the supply bandwidth required by motherboard with critically damped recovery.
• Stable feed back loop with phase margin to ensure stability when supplying into high di/dt and inductive loading.
• Good supply error feedback loop immunity to HF noise entering the supply from the PCs rail.
• Very good load regulation performance.
• Adequate PSRR.
• Very low ripple + noise. 

(I posted a few sensible figures in earlier post in the thread that people should look out for in the cases they are published by power supply manufactures)

 

 

 

As background to the above.

 

I designed and developed an advanced server supply system over the last few years. The power system was built as a technology test bed and development platform. It was designed specifically for flexibility to incorporate linear, switched buck and unregulated technology into the servers power system in any combination in order to look for optimum power arrangements. The aim was to test supply technology and topologies to makes choices based on experimentation and performance. For the last 3 years focus has been exclusively on how to supply single and dual core c621 based servers with up to CPU 20 cores per socket. So the "high current" types of system. The power system is a micro processor controlled and is "plug and play" so different designed linear power PCBs, third party DC-DC ATXs and unregulated choke supplies and "audiophile" grade SMTP supplies can be used in combination to supply the 3v, 5v, 5vSB, 12v ATX and 12V EPS rails of a c621 mother board.

 

With all of this work and the ability to select any of the supply technologies above in the final version of the server supply, what does the finished fully developed c621 supply system us ? There was absolutely not question what so ever, for sound quality, very high performance linear supplies (same for single and dual processor c621 systems). I need to add a proviso that the direct to board linear everywhere approach was very hard to perfect with the sound quality performance level I was aiming for. This in mind I accept and am not at all surprised when I read reports that linear everywhere results might not tally in other systems.

 

OAudio.  

 

 

 

 

 

 

   

 

 

       

 

      

[OAudio]

2 hours ago, matthias said:

There are two great posts from Taiko Audio about the topic LPS vs. SMPS:

 

https://www.whatsbestforum.com/threads/taiko-audio-sgm-extreme-the-crème-de-la-crème.27433/page-276#post-696870

 

https://www.whatsbestforum.com/threads/taiko-audio-sgm-extreme-the-crème-de-la-crème.27433/page-277#post-696871

 

Matt

 

Hi Matt,

 

I am defiantly familiar with these two posts, I spotted them some time ago and to be honest they contributed to why I said this.

  

9 hours ago, OAudio said:

Important to remember first that there isn't anything mystical about c621 boards, they are just incrementally slightly larger then previous Xeon architectures. Threads on other web sites make it sound complex but they are just computer PCBs. These boards have finite and quantifiable load requirements.

 

There are some interesting points made in these two links but there a few tangents and assumptions that appear to be there, in order to make the argument that linear supplies can not supply the current levels needed by a music server without descending into mush :-).

 

This explanation is lifted below for instance which is just a nonsense argument. The math is correct, but no one would be stupid enough to apply a linear supply technology to supply 100 watts at CPU core voltage of 1volts ! Its irrelevant to any real world power solution for a music server (certainly one based on any commercially available motherboard). Nobody will every [be able to] try this for a simple reason. Motherboards have and will "always" use buck technology to step down their external supply rail voltages to the high current low voltages supplied that enable high frequency power efficient chip designs. They are required to meet industry efficiency standards and the drive in data centres to reduce power consumption puts further competitive pressure on motherboard, chipset and processors designs to do this. However the point and all the numbers quoted are then used to underline a notion that linear supplies simply cannot meet the high current demands and perform to the specifications needed for class leading sound quality. Just not the case.....

 

"As an extreme example, let's look at powering a CPU. A CPU typically operates at 0.6-1.2 volts but can easily draw 100 watts or more, let's assume 1 volt for easier calculations. For 100 watts at 1V you are talking about 100 Amps(!) of current. As current drops voltage over resistance (Ohm's law), a CPU is typically supplied with a 12V voltage rail, so we need to regulate 12V down to 1V. If you would use linear regulation, and you'd have a 100 Amp current draw at 1V, you would also have a 100 Amp current draw at 12V (simplified), meaning 12V*100A=1200 watts. Then we need to feed this 12V by a 16-19V supply (again to account for voltage drop caused by current over resistance), let's assume 19V as that's a very common value to ensure broad compatibility, and we are talking 19V*100A=a shocking 1900 watts. The conversion efficiency here would be ~5%, the other 95% will just be converted to heat. A switch mode regulator is the inverse of this, it can regulate 19V down to 1V at a 95% efficiency wasting only 5% as heat. That is a 1900W versus 105W of power consumption."

 

Buck and linear designs both have their challenges. They need engineered solutions to manage their respective design hot spots but for my these two posts far too quickly quickly adopt the message linear + high current = guaranteed noise and buck can perform better despite their many many challenges. 

 

Put and H probe anywhere near any DC ATX and it just sings with emissions. Even more relevant put a scope or analyser on its output and look at the characteristic of the switching noise breakthrough. Both these tests should really be done in a meaning full way, by which I mean whilst driving the intended load, a mother board's power rails. These spit back wide band RF and transients into the supply in question. Its really down to how a supply of any technology behaves under this condition (not necessarily in to a constant bench load) that will determine what the sound quality will be.  

 

I wanted to link another post, but just could not find the particular post in the long thread. In it linear and buck conversion are again being compared in the same vain. Some FFTs are posted for an example buck supply and linear supply. The buck predictably looks like it performs well al least under constant load and this is attributed in part to an unusually high frequency switch loop in the design (I am familiar with the device I think is being used which switches around 2MHz). The FFT of the linear supply that is provided in the post is however dreadful. I would hang my head in shame if a linear with that spectral content were being used in a supply here in the modules I make and these will supply 24 amps per rail and give the electrical characteristics I have mentioned in earlier posts in this thread.

 

The point is for me is that the posts maybe not be intentionally loaded to paint buck technology as the logic conclusion to the design decisions that have to be made, but I cannot follow the arguments and logic stated and accept that linear supplies are an inferior choice. I have some direct experience but to those that don't it all looks and sounds very complicated and credible. 

 

OAudio

    

 

[OAudio]

Matt hi,

 

Perhaps I should offer some additional information that might help understand why I have a least a little confidence that a linear everywhere to the motherboard supply system of the right performance level can defiantly hold its own. Before the lock down my server with its power system was in a very high end audio system for some fun back to back comparison with a Taiko extreme and another leading server. Using this as a reference point I am reasonably sure that unless the Taiko DC-ATX generates sound quality of a standard greater than their premium extreme server product, that full linear to the motherboard supplies of the correct standard will compare favourably with the new ATX.

 

I am honestly open about this however and you never absolutely "know" until you can get both in the same system and setup.

 

Regards,

     

[OAudio]

14 hours ago, Nenon said:

The unregulated LPS I am currently using has about 1.4mOhm output impedance (it also has about 5uV ripple).

 

@Nenon 

 

Wow, 5uV = ~-135db of rejection ( assuming a ~30v rail ) ! 

 

I have been playing along in the background for a while using simulations of an unregulated supply based on the component choices in your earlier posts. With components listed and tweaking components parameters to near ideal performance levels its hard to get <3mv pk - pk @ 1 amp constant load.    

 

5uV is exceptional. Can I ask how you measuring the ripple and and under what test conditions ?

 

Thanks. 

[OAudio]

On 5/7/2021 at 6:46 PM, Nenon said:

On 5/7/2021 at 4:14 PM, OAudio said:

 

@Nenon 

 

Wow, 5uV = ~-135db of rejection ( assuming a ~30v rail ) ! 

 

I have been playing along in the background for a while using simulations of an unregulated supply based on the component choices in your earlier posts. With components listed and tweaking components parameters to near ideal performance levels its hard to get <3mv pk - pk @ 1 amp constant load.    

 

5uV is exceptional. Can I ask how you measuring the ripple and and under what test conditions ?

 

Thanks. 

Expand  

Have you tried simulating the dual choke CLCLC configuration? 

 

 

 

I had not spotted the move to CLCLC. Modelling that configuration closes the gap somewhat :-)