Building a DIY Music Server

[Nenon]

2 minutes ago, Downtheline said:

I think you have also used the pink faun i2s bridge in some of your builds. I may use an AMD CPU to be able to use the i2s bridge, but prefer the intel i9-9900k with the maximus gene XI mobo... pink faun says some intel CPUs work with their bridge, some don't. Do you know first hand if these play well together? Anyone else care to chime in?

Thank you again

 

I do. the PinkFaun I2S bridge would NOT work on the i9-9900k with the maximus gene XI mobo. That's one of the reasons I went the AMD route some time ago. 

[Nenon]

I need to practise pipe bending more (definitely not my skill), but here is my first attempt to passively cool two Xeon CPUs in the ASUS SAGE motherboard used by Taiko Extreme.

 

I know it looks ugly but it works. In fact the longer ugly looking pipes on the left give me 4 degrees lower temperature than the stock HDPlex pipes on the top right. Perhaps because they cover more surface area? Or they are better quality? Or both? 

 

Still a lot of work to do on this build... it's not ready if you are planning to ask. 

[Nenon]

1 minute ago, Downtheline said:

Interesting.  You doubled up the hdplex heat sink cpu mounts and seem to have added some generic heat sinks to the uncovered edges of each cpu. How do you attach the heat sink cpu mounts to the socket?  2 screws each? 

 

I was just told the case was incompatible with the socket! This looks great!

 

The case is indeed incompatible with those Xeon CPUs. The LGA3647 socket has a completely different mounting mechanism - four screws. I custom made my own cooling kit. Had to use the cooling kits of four HDPlex fanless cases. I took a LGA3647 fan mounting kit and modified it. Took some pictures during the process and will post them when I have more time. 

[Nenon]

5 hours ago, Alexis said:

The 1^st question is what will be better: PinkFaun USB Bridge + SOTM or PinkFaun SPDIF Bridge?

That depends on your DAC and whether it sounds better with USB or SPDIF. 

 

5 hours ago, Alexis said:

The 2^nd question is AMD (Ryzen 5) or Intel (Core i5) platform? Euphony make its servers on Intel, PinkFaun on AMD. I understand that in terms of SQ it’s reasonable to rely on personal preferences (and here my choice is more likely to be a smoother AMD), but what do you recommend in terms of better compatibility with PinkFaun Bridge and Euphony?

In terms of compatibility with the PinkFaun SPDIF bridge, you would have to use AMD. It won't work with the Core i5.

PinkFaun USB works fine with both. Euphony also works fine with both. 

Even CPUs under the same series (i.e. Ryzen 5 or Core i5) can sound different. 

 

5 hours ago, Alexis said:

Would CPU without a Graphics Processor be preferable?

It's one of many factors to consider. I would not say that generally speaking CPUs without GPU sound better. I've heard CPUs with GPUs that sound better than CPUs without. I've also heard the opposite. 

 

There are several documented good builds on this thread and this forum. You can try one of them. It depends on your budget of course. 

The Intel NUC7i7DNBE is a good start. The builds in this thread are a significant step up from the NUC.

BTW, someone compared the NUCs to a BMW and the builds here to a Rolls Royce. That's far from true in my opinion. I would not give the NUC a BMW status. In the best case it's more like a Honda Civic in my books. 

Whatever you do, invest the most on power. It's the most important part. I typically invest a lot more (at least double) on the power supply than the computer parts. 

[Nenon]

On 5/4/2020 at 10:53 PM, Downtheline said:

Have you tried the pink faun i2s bridge yet?

Yes. It works fine with those CPUs.

 

4 hours ago, sig8 said:

The Xeon based server you are making; would it be able to do HQP DSD64 to DSD512?

You have more cores (a lot more) but lower clock speed. I would actually go with AMD for HQP. No need for HQP in my system, so this build would be tweaked for high CPU/process isolation. 

 

Speaking of high CPU/process isolation, I am also planning to try Windows 10 Enterprise LTSC 2019 instead of Euphony. I actually just happened to have access to the IOT version as well, which is what the Extreme uses. Since I don't have root access to Euphony, it is hard to optimise everything properly with the limited Expert Settings in Euphony. Windows would give me a much finer level of control. It would be interesting to try Windows again after I have been a Euphony fan for a long time. Tweaking this server properly takes a lot of time. But I think it's worth it. I also needed a really good power supply for this build. But I will share more info on that some other time. 

My custom industrial RAM is due at the end of the month. Quite excited about that. 

[Nenon]

 

On 5/10/2020 at 3:32 PM, ASRMichael said:

Moving to the Dark Side of the force!!!!! 😂

 

Haha, yes. Euphony does not sound as good on the Extreme as the custom built and tweaked Windows. Time to investigate.

 

On 5/10/2020 at 3:32 PM, ASRMichael said:

What tweaking are you planning on doing? maybe Euphony can give me tweaking options?

We'll see - mainly CPU isolation, affinities, and priorities for now. Also, custom asio drivers. 

 

20 hours ago, Hauser said:

A good deal of time spent on Google rewarded me with info on another server; Phasure's Stealth 3. 

Looks interesting but I see a lot of fans inside. It's easy to built a powerful server with fans. And it can sound good. But if you want to squeeze out the last bits of performance, a fanless approach is the way to go. It also has a big transformer inside the chassis. Every time I have put a transformer close to my motherboard I hear sound quality degradation. I've done that test many times while playing and it's quite illuminating. That's why all my builds have two chassis. Of course everything is a compromise - in my case I needed to add connectors and umbilical cords, but with attention to details and good materials, it's the better compromise. 

But I don't want to make any conclusions for the Phasure's Stealth 3. I've never seen or heard one. However, it does look like a good option for the Hqplayer folks.

 

16 hours ago, dminches said:

My biggest roadblock now is that I would love to have my files loaded on a card inside the server and not pull them from my NAS.  The issue is the amount of music I have.

3 hours ago, Chopin75 said:

I have similar problem

My Euphony/Stylus based servers that run from RAM and cache everything before playing sound really good from NAS. 

My recent custom Windows 10 Enterprise LTSC / Roon attempt does not sound good playing the same files from the same NAS. It shines with files on an Optane card. I am also looking at some NVME storage now - thinking between Intel 660p, Samsung 970 PRO, and some industrial wide range temp Apacer ones. 

 

 

6 hours ago, Hauser said:

I thought I'd try using the bottom of the case for all the supplies; the lower 50mm, build a second level for the motherboard; the top 90mm.  I will probably use MP Audio regulators and 4 or 5 Toroidy transformers. 

4 or 5 Toroidy transformers right under the motherboard is not a good idea from my past experience. The test I mentioned was done with one of their Supreme transformers, which is enclosed in a case and epoxy filled. The closer it gets to the motherboard, the worse it sounded. I've had much better results with 2 chassis. 

 

20 hours ago, Hauser said:

It would be interesting to know how Nenon's custom RAM performs.

Me too! Can't wait for it.

 

18 hours ago, guiltyboxswapper said:

The real shocker is the fact I can run HQPlayer from the same "noisy" machine and it still exceeds the opticalRendu and not in a small way at all. 

There is no comparison between the Sonore products and the custom DIY servers we are building. But our hand built servers are quite more expensive too. 

 

7 hours ago, Hauser said:

Having looked at the Asus motherboard Taiko use, the ethernet connections channel through the board chip.  You may get better results connecting with a card through PCIE to the CPU.

I think there is a general (mis)concept that USB and ethernet ports that are not CPU-direct sound worse than CPU-direct. While this may be the most common case, it is not ALWAYS the case. Taiko Extreme's USB port goes through the ASM controller, and it sounds really really good - it's actually very hard to beat. My Asus ROG Crosshair build with ultraOCXO chipset clock is another example - after the clock upgrade the chipset USB ports started to sound significantly better than the direct CPU USB ports. Also, when you add a PCIe card to a PCIe slot that is CPU-direct, the card has chips on it as well, so you are not bypassing the controller chips. I know you referred to ethernet NICs, but they also have controllers. The Extreme comes with a Startech PEX1000SFP2 PCIe card, and many owners prefer that.

 

[Nenon]

Asus SAGE / dual Xeon CPU passive cooling

 

Below is what I did to passively cool the dual Intel Xeon Silver 4210 CPUs in the Asus SAGE motherboard. 

 

I could not find a solution to passively cool the Intel Xeon Silver 4210 CPUs. Streacom, HDPlex, and some of the other common passive computer chassis do not provide a solution for LGA3647 sockets. Also, most of them are limited to 65W TDP or 95W TDP. Cooling down two 85W TDP CPUs is quite challenging. Turemetal UP10 looks like a really nice case, but the ASUS Sage motherboard (12'' x 13'') does not fit inside according to their specs. And even if it could fit, the mounting on the LGA3647 sockets is very unique. You need very precise parts to be able to mount the CPU without damaging anything. 

 

The first thing I did to get started was to buy 2 x Noctua NH-D9 DX-3647 4U CPU coolers. Those are active CPU coolers, but I wanted to make sure everything was working properly. The motherboard posted, both CPU tested well, so I started looking for a way to do the passive cooling. 

 

The LGA3647 comes in two variations - narrow and square. The Asus SAGE motherboard uses the square version. After some research, ordering some parts, returning some, I decided to use the Dynatron B9 CPU cooler as a base. This is how it looks - top and bottom:

 

 

We don't need the fan, so that can be removed. But we need the mounting mechanism for the socket. After removing the fan, we end up with a heatsink that can be used as a base.

 

The surface area on those Xeon Silver CPUs is much larger than a typical consumer CPU such as Intel Core or AMD Ryzen. I would need two HDPlex passive cooling kits to cover one CPU. I ended up getting four HDPlex H5 chassis for this project. 

 

An explanation of the process with pictures follows below.

 

I started removing some of the material to make space for the HDPlex cooling kit.

 

Now we have enough space for the copper HDPlex cooling block.

 

We need the surface to be as smooth as possible. Sanded with 400 grit sandpaper, followed by 600 grit, 1000 grit, and 2000 grit. I used wet sandpaper from an auto parts store and some soap water. Here is the result.

 

I did a little more fine sanding and polishing to prepare the surface. It's now ready to install the 2 HDPlex copper blocks. They fit perfectly. My research time was well worth, and the Dynatron B9 looks like the perfect solution. 

 

The next big research was on thermal epoxy. I needed to glue the two HDPlex copper blocks to the B9 with glue that would transfer the heat from one material to another as efficiently as possible. I picked the MG Chemicals 834HTC-A High Thermal Conductivity Epoxy for that. 

 

Cleaned the surface with 90% alcohol, let it dry, and applied a thin layer of the epoxy. I used two heavy duty clamps and let the epoxy cure in my oven on a low temperature for a few hours. All done with that part. Here is the final result. 

 

I repeated the same process for the second CPU. With two in place, it's time to install them. 

 

Here comes the second problem. No passive cooled chassis is designed to cool two CPUs. The HDPlex H5 (and most others) has two heatsinks, and typically only one is actually used for CPU cooling. The idea is to use one heatsink for each CPU. In order to do that, I had to buy new cooling pipes, a pipe bending tool, and learn how to bend them. I had to do some reading on cooling pipes, learn how they work, learn about the different designs, different materials, etc. I did not know any of that stuff before this project. 

There are quite a few things to consider - the shape, the materials, the quality, how you bend them, etc. They are filled with liquid and you can't cut them. Also, you have to be careful not to crack them when you bend them. The bending radius can impact the performance. They come in different lengths and some are better quality than others. I liked the products a company called "Advanced Thermal Solutions" makes.

 

Bending pipes is a skill that I need to practice more. Here is my first attempt - looks ugly but it worked great.

 

The CPU on the top uses the stock HDPlex cooling pipes. The CPU on the bottom with the six longer ugly-bended pipes is what I did. 

To my surprise when I turned on the computer, I realized that the bottom CPU's temperature is lower than the top CPU. One was in the low 40's after a couple of days of playing music and the other was in the high 40's. That's degrees Celsius obviously. We have to be careful with a dual CPU configuration, because one CPU could be hotter because it is doing more than the other. I made sure that was not why the top CPU was a few degrees hotter. 

 

I decided to replace the stock HDPlex cooling pipes with new ones. I did much better job with the bending, but I don't have a picture handy. You will have to trust me on that one :). 

 

One problem with the HDPlex cooling kit is that the pipes are short and don't cover the entire cooling block. I'll refer to this post for more info / picture:

On 2/28/2020 at 2:18 AM, StreamFidelity said:

The HDPLEX package included quite valuable copper bars and blocks for passive cooling.

 

 

By replacing the stock cooling pipes I had the chance to use longer pipes and cover more surface. 

Not sure if the cooling pipes I used are better than the stock HDPlex or it was because they covered more surface area, but I saw about 7-8 degrees lower temperature with the new pipes. I am guessing it's both - better quality pipes and more surface area. 

In fact, now the top CPU is about 1-2 degrees cooler than the bottom CPU. 

 

The thermal epoxy also takes about a week to 10 days to settle completely. It gets more efficient over time. Overall quite happy with the result. CPUs stay in the 40's depending on room temperature. It is 84F degrees in Chicago today, and I saw them running as hot as 49C. I'll go deal with my A/C now that I am done with this post. 

[Nenon]

2 hours ago, Peter Avgeris said:

Just to add my two cents. I am a mechanical engineer with too much experience in metal works. The idea of having a copper base (Dynayron) and on top two copper bases is far from ideal. You should have made two new copper bases on the milling machine with mounting mechanism for 3647 square ILM socket. It is so simple. And glueing the top two copper plates does not make sense in the engineering language. All the rest seems to be really well executed. The final judge is the final result and it seems to be good. Congrats. 

 

It did not feel like I could detach the mounting mechanism from the Dynatron and attach it to another copper base. I was looking to do something like that originally. But it looked quite complex, and I did not want to ruin CPUs and motherboard for $2200. It did look simple at first, but the devil is in the details... 

The thermal epoxy was a crazy idea of mine, really a shot in the dark... I thought it would be a piece of crap at the end but it turned out pretty good. 

 

The Taiko Extreme CPUs with their much more sophisticated cooling mechanism are running at low 40s. Their heatsink is giant compared to the HDPlex. It's also made from pure copper. And if you look carefully at the chassis design, the whole chassis acts like a big heatsink. That's not the case with the HDPlex case, but I get some benefit of utilizing the heatsinks on both sides. 

 

Having my server running just a couple degrees hotter than the Taiko Extreme with this primitive, almost no tools required, DIY method is a pretty good achievement. I am sure there are many ways to make this better. But since it was that simple, I decided to share as other people may want to try it. 

 

I thought the thermal epoxy and the connection between the two copper pieces would be a huge bottleneck. But it seems like the epoxy does a pretty good job. It almost feels like the HDPlex case is the bottleneck at this point. 

 

Having said all that, if anyone has a better solution to offer, please let me know. The cooler we can keep our servers, the better. So even if something would bring down the CPU temperature by a couple degrees, I would be willing to try it. I am also considering a DIY chassis with better (bigger) heatsinks that can dissipate more heat. Not ready to go there yet and not sure if it's even needed. But it's probably not that complex to take a pre-made (i.e. power amp) chassis and mill a few grooves for the cooling pipes. 

 

[Nenon]

3 hours ago, Brianfromspace said:

It made me stop buying a Innuos server and after reading all 17 pages of this thread yesterday night,

Innuos is a great option if you are not into DIY. 

 

2 hours ago, Brianfromspace said:

-I understand the other soldering you did was to be able to power the clock/CPU and the 800W from the DC3 unit you have. Are the cables you use also available on the market pre build? If not: are you willing to build and sell me a similar setup (without the one for the clock? It looks awesome and again I don't dare to start soldering on this server myself.

(The only option I see w/o soldering is the HDplex 400W ATW LPS, I see a separate connector for the CPU, but I don't know if that means it's powered separately. If your answer to the last question is: no can do, what would you advise me as a power supply for the board/CPU/JcatUSBcard)

Ping me on PM on that. I occasionally do some DC cables for members here but don't always have the time.

 

2 hours ago, Brianfromspace said:

-Could you tell us the specs on the DC3 unit that makes it ready for this setup?

21V - 5A for the HDPlex 800W DC to DC ATX

12V - 10A for the CPU / EPS

5V - 3A for the JCAT USB XE

5V - 3A for the JCAT Net Femto

5V - 3A for PinkFaun ultra OCXO clock

 

2 hours ago, Brianfromspace said:

Why didn't you use a fibre PCI card instead of the Jcat Ethernet to eliminate noise from ethernet cables? I will only stream from tidal/roon/EuphonicOS, i'm wondering also how much it matters making the internet input clean.

I prefer the JCAT Net Femto in this build. If has better clock and external power. 

Fiber NICs, such as the Startech PEX1000SFP2 card draw power from the PCIe slot, which generates more electrical noise than an externally powered JCAT NICs. I have found that tweaking my network and adding fiber elsewhere is a better approach and works better for me. Tweaking the network is a whole other topic.

 

2 hours ago, Brianfromspace said:

-Why do you use only 8gb or RAM? When booting Euphony from RAM, is this enough?

Yes, it's more than enough in my case. Even 4GB works, but you are too close to the edge.

[Nenon]

32 minutes ago, Peter Avgeris said:

First of all, I really can't see any reason for building such monsters for music playback. Music always sounds better with lower power machines, provided the computational power is adequate. Exceptions? Of course: ROON playback software. During the years, it requires more and more power to work properly. More cores but this comes at an expense for sound quality. With an Euphony system, or even JPlay server, I really can't see any point for going that extreme. My Euphony systems never sounded better than what they do now, with only 2 Xeon cores for server and another 2 for the Endpoint (4-core machines, no hyperthreading, no turbo boost, 2 cores suspended from BIOS, no VGA, no SATA/SAS). 8GB of RAM per machine.

 

Roon is another story. But again, if you want that much computational power, you can go for it. 

 

 

@Peter Avgeris It seems like you have no idea how much you are missing and how much better your digital source can be, but I don't have time or the desire to argue about this. If you are happy with your source that is all that matters. 

 

 

9 minutes ago, Peter Avgeris said:

Yes if you are located in Europe. No if you are in USA.

I don't want to be drawn into such a terrible trouble with all customs related issues for overseas shipments. For European affairs, yes, I can do it. It is a piece of cake, don't forget this!

Some members here (including me) can definitely use your mechanical engineering skills and take you on this offer. I suggest we start a new thread to discuss the options. There are a couple of people in Europe that already have the Sage motherboard and are looking for a passive cooling solution. I will refer them to the new thread. Perhaps someone can send you a motherboard, you can do a prototype, and we can do a group buy from you. Thank you for offering this. It's really good that someone can do a more professional heatsink than my ugly mockup. For those of us in the US, I am sure our European friends will help us. You can ship to them, and they can handle the rest (distribute to the US).

 

 

32 minutes ago, Peter Avgeris said:

I can provide you with any pieces you want, specially made for HDPlex chassis and Advanced Thermal Solutions 10mm copper heat pipes design, specially made for HDPlex chassis.

BTW, the Hdplex uses 6mm pipes. 

[Nenon]

6 minutes ago, Peter Avgeris said:

Really? I am losing so much! WoW! This is very interesting. I am all ears. I do not consider myself a digital guru. I admit that I know quite a lot of things for designing tube amplifiers with exotic transformers, all designed, wound and manufactured in-house, but I am mostly easy going with digital. I tried to use high power streamers but they never gave me that extra magical hint. Whatever I listen through any sources, specially digital, is more or less the same to my ears, as I have been used to the rather extreme differences between power amplifiers and crazy, out-of-this-world field coil loudspeakers, with exotic heat treating procedures for the poles. If you could assist me on this, I would be more than happy 🙂

 

Well, completely offtopic, but since you opened up the subject, here is what I have been working on while waiting for my custom Apacer memory. 

 

This is an Extreme, cost no object, version of the Audio Mirror Reflection 45W Set monoblocks with custom transformers, Mundorf Mtube and Mlytic capacitors in the power supply, Z-foil resistors all over, Duelund CAST PIO tinned copper capacitors, exotic wiring, WBT Nextgen connectors, Furutech NCF inlet, etc. The list goes on. I use four of these to bi-amp my high efficient speakers, which have 18'' woofers. I tried Jeff Rowland 625 S2 amps for the woofers thinking that might be the ultimate bass amplification. But those SET monoblocks outperformed the Jeff Rowland by a huge margin in the bass section. It's unbelievable how much more detailed and vibrant bass I got with those. That was one of those shocking and completely unexpected moments. I was so biased towards the Jeff Rowland that it took me quite some time to believe it. But it sounds like this is a small toy compared the stuff you are talking about :). 

 

Digital, on the other hand, is completely different. You really have to tune your listening differently to identify the small differences in the digital domain. Needless to say, the power supply is the most important component in the digital source. And the power supplies that do miracles for analog gear don't necessarily work that good for digital. A high powered server with not so optimal power supply can sound pretty bad. But once you nail down the power, the magic starts. 

 

The Asus ROG Crosshair VII Hero build mentioned earlier in this thread with the upgraded motherboard ultraOCXO clock sounds amazing. This dual Xeon Asus SAGE will surpass the Asus ROG build but it takes a lot more effort. One is running Euphony, the other one is running Windows 10 Enterprise LTSC. They are tuned completely different. One sounds better playing music from a NAS. The other does not sound good from NAS and needs local storage. It also takes quite a bit of effort to tweak your network. But when it all clicks, it's magical. 

 

47 minutes ago, Peter Avgeris said:

Go ahead, start a new thread and invite me there.

Perhaps one of the people who are interested here can take case of this? If there is no volunteer, I can do it later this week.

 

48 minutes ago, Peter Avgeris said:

I have a friend in Germany, Heidelberg area, he has this board with dual Xeons, dunno if he could send it to me.

If this is Dirk, that's one of the people I had in mind as well. Haven't talked to him lately, but I believe he would be willing to send you the motherboard. 

 

 

[Nenon]

7 hours ago, ray-dude said:

 

...and this is EXACTLY why I'm in complete denial about the existence of SET monoblocks...something like this in the house would become a full time job (obsession?)  Gorgeous work Nenon!

 

LOL. Ray - I was thinking about you when I posted that (but there was no option to hide my post from you).

No, forget about SET. It's a horrible thing. The tubes awfully distort the sound. You have to do a lot of tube rolling to find a tube with less distortion (with limited crackling and hum). The tubes need 500 hours to break-in and die at around the 510 hour mark. If you are even that lucky. Sometimes they die at the 495 hour mark, just 5 hours before the burn-in process is completed. A big percentage of the tubes just explode - it's very dangerous. 

Note: this message is just for @ray-dude. Please ignore if you are not @ray-dude 

[Nenon]

16 hours ago, Mr Morris said:

Can I ask a basic question? Apologies if it is answered elsewhere, but I have been inspired by this thread and put in an order to  Sean Jacobs! My question relates to the motherboard wiring. I was intending to wire all common voltage pins together, but notice you have not used them all. Is this because you are not using a SSD and so wiring the alternative separately? Thanks 

 

I have done it both ways. Some cheap ATX cables are so thin that two wires (plus and minus/ground) cannot provide the necessary current. Doubling those wires is quite necessary. If you use better (and thicker) wires, like the ones you see here, and your motherboard does not draw a lot of current, you don't need all. Having said that, running multiple wires to each voltage is a good idea. Go for it. 

All 3 voltages on the ATX combined together need less than 30W. That's why it did not really matter that much. If I was adding PCIe cards such as powerful video cards that need a lot of power, it would have been a different story. But the PCIe cards in this case are externally powered and there is nothing else connected to the motherboard (besides the small m.2 Optane card of course). The CPU/EPS is where most of the current draw happens. And I use all 8 wires there. 

Good luck with your build. Feel free to post some pictures and feedback here when done. I think most people reading this threat enjoy looking at other people's DIY builds. 

[Nenon]

2 minutes ago, Mr Morris said:

I didn't realise that the grounds are all common. I wanted to keep them separate back to the caps. Or Do different ground pins correspond with particular voltages?

The entire ground plane is common on every motherboard I have seen. 

[Nenon]

1 hour ago, ASRMichael said:

@Nenon I see you are using HDPlex 800w card again in your build. What benefits are you hearing versus direct connection from SJ LPS? 

 

I keep going back and forth. But I mostly use the Hdplex for convenience. 

 

My builds typically need 6 rails:

- 3 rails for the ATX

- 1 rail (high current) for the CPU/EPS

- 1 rail the JCAT Net Femto NIC

- 1 rail for the digital cart (PinkFain or JCAT)

 

You can either go 6 rails or you can do 4 high quality rails (like @adamaley did with his DC4) that would be:

- 1 rail (high voltage) for the HDplex 800W DC-ATX --> feeds the 24-pin ATX connector only

- 1 rail (high current) for the CPU/EPS

- 1 rail the JCAT Net Femto NIC

- 1 rail for the digital cart (PinkFain or JCAT)

 

I can go either way, and I enjoy both. There is something that I like about the 800W DC-ATX when fed by really really good LPS (otherwise I don't). For the ultimate system I would go with 6 high-quality rails. But the 4-rail system is nearly as enjoyable as well. 

 

The biggest difference makes the power on the EPS connector and the USB card. Obviously if you feed a Hdplex that feeds 3 rails, you want to have very clean power there as well. 

[Nenon]

1 hour ago, novaca said:

 

Do you think that PH SR4(T)-19 (2A) for the HDplex 800W DC-ATX and PH SR5(T)-12 (6A) for CPU/EPS (Xeon E-2136) would suffice?
Or should (I have to) think of PH SR7?

 

38 Watts for the ATX and 72 Watts for the EPS would probably work.

 

Around 30W is what most of my motherboards needed to power up (that is 30W feeding the Hdplex 800W DC-ATX). 38W would not give you much headroom for things to power up through the motherboard (i.e. PCIe cards without external power supply, NVME storage, SATA devices, etc.). But it would work with most barebone motherboards from my experience. 

 

72W for the EPS would also probably work if you are not using a lot of processing power. But CPUs are very fast in current demand. Your average current may be something like 30-40W but CPUs tend to create those microburst of power demand that may be higher than double what this LPS can provide. While it would still work, you might be restricting the dynamics. I tend to go with heavily oversized power supplies, especially for the EPS, and that has always sounded better in my system. 

[Nenon]

Some updates from me below.

 

1. I completed my tube SET monoblocks a couple of weeks ago. They sound amazing! Even better than they look. The Duelund Cast tinned copper caps are still breaking-in as well as all the Mundorf MTube and MLytic AG capacitors I used but they are 99% there. Add to that, the custom silver wires used throughout, Mundorf silver/gold wire for the power supply, custom transformers, custom chokes, Z-foil resistors, WBT silver connectors, NCF IEC inlets, Grade A Sophia Electric tubes, Grade A Shuguang WE6SN7 PLUS tubes, SR Orange fuses, etc. I used to love my Lundahl output transformers but those beat the shit out of the Lundahls! These monoblocks are something special. They really are a piece of art, and I am very proud of the end result. Sorry for the ugly photo, but that's the best I have at the moment.

 

2. I am still playing with the Asus Sage build, trying different Optane cards and NVME storage (in RAID0 VROC configuration). It seems like no NAS will be used for this build. It's interesting because the NAS sounded best in my previous builds but not here. 

 

3. The custom Apacer RDIMM memory should be ready next week. Looking forward to try it on the Asus Sage motherboard. Speaking of Apacer, they discontinued the new RAM models that used the 2666MHz Samsung chips as Samsung stopped manufacturing some parts. That's quite unfortunate, as that was the best sounding memory. Here is a list of the older Apacer models (still using the top grade Samsung RAM and still wide range temperature but 2400MHz):

ECC DDR4 DIMM 4GB: 75.B93GZ.G000B (recommended for most server builds)

ECC DDR4 DIMM 8GB: 75.CA4GZ.G000B

NUC DDR4 SODIMM: 75.B93GJ.G010B

Hope that helps to those who have not upgraded to Apacer RAM yet. 

If you can still find it somewhere the ECC DDR4 DIMM 4GB model # D31.23185S.001 is in my opinion the best sounding RAM. Mouser still shows some on backorder, but I highly doubt those backorders would be fulfilled at all.  

Last note: I did not list non-ECC models, because ECC works (as non-ECC) on most motherboards and sounds better to my ears.

 

4. I managed to obtain a number of Buffalo BS-GS2016 switches before they disappeared from the market. Some will be used in my system. Some are promised to friends. And I might have some left that I would not need when I finish testing. 

After the excellent post by @auricgoldfinger about his modified Melco switch with PinkFaun ultraOCXO clock, I am determined to try that on the Buffalo as well. With some luck, my next switch would have the following features:

- Modded Buffalo BS-GS2016

- Replaced clock with PinkFaun ultraOCXO powered by a separate 5V rail.

- Two-rail Sean Jacobs DC4 LPS powering the switch and clock.

- Chassis vibration damping.

- Gaia isolation feet.

- Possibly a bank of capacitors on the input similar to what Melco does (to be determine if that's an improvement). 

 

5. Not really an update, but I thought the following post might be more useful here:

On 6/2/2020 at 12:12 PM, Nenon said:

@ASRMichael

I suggest we leave the power supply outside of the box. 

 

The transformer impacts the sounds in two ways - it generates a lot of EMI and vibrations. 

Putting a transformer inside the motherboard chassis always degraded the sound quality in my system. I tried this many times and the result was always the same. Actually much more audible than I thought. Even while playing, moving the power supply closer to the motherboard has an audible impact. Based on multiple tests, with different transformers but mainly the Toroidy Supreme, I decided I would never host the transformer in the computer chassis.

Somebody was preaching that this is the best way to go in the other thread. I can't disagree more but did not want to argue with him. 

Adding (good quality) connectors and umbilicals increase the output impedance of the power supply, but the impact of that impedance increase is nowhere near the damage the transformer does when in proximity to the motherboard. 

 

Taiko uses a thick plate (looks at least 5 mm, maybe more) and a thick piece of PANZERHOLZ wood to isolate the transformer. That provides both EMI and vibration treatment. I did try a 3 mm plate between the transformer and the motherboard as well as a 25mm heatsink. But I could always hear sound quality degradations to some extent. Positioning the transformer changes the level of impact. With a lot of testing you can find a position (i.e. vertical or horizontal and at a specific angle) where the transformer would have less impact. But you need to do that in a prototype case and then design the case accordingly. Starting with the final version of the case without having the option to make changes after careful testing is not the best way to go if you are to place a transformer inside. That position of the transformer and a thick plate would help with EMI. Vibration isolation also helps. I have even tried putting the transformer on a set of Stillpoint Ultra 6 feet. But I could never get it to the point where there is no impact if the transformer is inside the chassis.

I feel like placing the transformer inside the Extreme was a carefully measured compromise. The Extreme chassis is the most expensive part of the server. If they had to make it in two chassis it would add a lot more to the cost and shipping. It would also make the server so big and heavy that it would not fit in many systems. My guess is with careful placement and isolation the overall result was satisfactory at the end. 

 

With the transformer outside the chassis, we can do other improvements. For example, if we are talking about a Sean Jacobs power supply, we could host the regulators inside the motherboard chassis and use super short cables for the ATX and EPS connectors. Yes, that is certainly a good idea, but we have another problem - heat. The cooler we keep the power supply and the computer components, the better. They would last longer and sound better. Let's take Sean's DC4 for example, his best LPS. For the amount of current my dual Xeon Asus Sage draws, a DC4 would have to dissipate at least 50W. That would keep the overall temperature of the Turemetal case several degrees higher. I don't think that's a good idea. So for a dual CPU configuration the right balance for me is to keep the power supply external and use good DC connectors and umbilicals. A 65W TDP single CPU system is another story - I would put the regulator inside fot best performance on such system. But a dual CPU build has its own set of challanges that need careful consideration. 

 

Having said all that, it might be a good idea to add space (and preferable predrilled holes) for two connectors in the chassis - one for the EPS and one for the Hdplex. 

Jaeger 536603006 is a good option - Paul Hynes uses them as an optional (XLR) upgrade to his custom dual regulated SR-7.

And if you want to go crazy, the Jaeger 533760006 is a monster overkill connector. I use it, and it's the best connector I have seen (don't often look at connectors, so there might be better ones). For those who don't want to use connectors, the two connector openings (holes) can be used to run cables through them. 

 

Last thing to say. Accounting for power supply components in the chassis would create a lot of challenges as I don't think we would find 10 people who would like to use the exact same LPS in the exact same configuration. I feel like it would make this project impossible to complete.

My recommendation is to stay away from that. Focus on the requirements we already agreed. And if people like my idea about connector holes, let's add that to the wishlist. You need 5 holes for the Jaeger connectors and they are not easy to do manually. 

 

That turned out to be a long post - sorry. 

 

Finally, the life is getting back to normal in Chicago, and I am planning to enjoy the summer. There will be less frequent updates from me in the next couple of months. But there will be some!

 

[Nenon]

18 minutes ago, elan120 said:

What would you recommend for power switch and fuse holder?

I bypassed the power switch completely. The fuse holders, as well as the chassis, and some other parts like their proprietary biasing system, protection board, etc. were provided to me by AudioMirror. These are AudioMirror Reflection 45W SET monoblocks at their cores. 

If you are looking for an oversized / overbuilt fuse holder (double check its not too big before you order), check those out:

https://www.digikey.com/product-detail/en/schurter-inc/0031.2303/486-1984-ND/641186

https://www.digikey.com/products/en?keywords=0031.2323

 

23 minutes ago, RickyV said:

wow Nenon really nice work those amps look amazing and judging from the components used they must sound amazing too. Those Duelund caps alone cost at least 220 euro apiece for the smallest one, hope you didn’t need to many of them.

Using 1 uF Duelunds... two per monoblock

 

24 minutes ago, RickyV said:

 I may have mist it before or it slipped my mind but what speakers do they drive?

I built The Loudspeaker (TL-1) by Troels Gravesen about 18 months ago, and that has been my reference since then. Needless to say, I went overboard with my OCD. They also use Duelund tinned copper caps as well as Jantzen Amber-Z Caps, Path Audio resistors (not a fan of the Duelund resistors), special wire, WBT connectors, top quality 1.5'' thick baltic birch, etc.

They have 18'' woofers. The original design was using DSP and ClassD Hypex plate amps for the 18'' woofers, but I am very sensitive on bass details and definition. After a lot of testing, I settled on passive crossovers and bi-amping. Tried quite a few amplification options, including the Jeff Rowland 625 S2 amps as mentioned in one of my previous posts. But when I heard what those AudioMirror monoblocks did, I could not go back to anything else. That was not a self-intuitive test result. Typically one of the weaknesses of tube amplification is deep base. But I guess 95dB efficient 18'' woofers are a different beast. 

So, here I am, bi-amping with two sets of monoblocks (4 monoblocks total)!!! The question is what am I going to do in the summer months with those heaters... it was nice and cozy in the winter :). 

 

1 hour ago, RickyV said:

Enjoy the summer sun but hopefully you do not pull a romaz on us. 😃

LOL :). The plan for now is not to buy an Extreme, but to work together with the community here, so we can come up with a DIY server that brings us closer to extremes!

BTW, one member here just ordered a PANZERHOLZ wood bottom plate for his HDPlex H5 chassis. I am excited to hear what that would do!

[Nenon]

2 hours ago, Ianderson said:

Nenon - You mentioned shooting higher than an Innuos Statement, so this unit would sound better? Curiously if you were to produce another, what would one of these cost?

Those are two difficult questions to answer. Unfortunately, I can't answer with Yes/No and a dollar amount, so you may not like my answer.  

 

There are actually 3 builds in this thread - the Intel Core i9-9900K I started the thread with, another build with a full size ATX ASUS ROG motherboard / AMD Ryzen 7 3700x / ultraOCXO clock on the motherboard, and now the dual Xeon CPU that uses similar hardware as the Taiko Extreme.

 

Those three builds explored 3 completely different CPUs - a consumer Intel, a consumer AMD, and server Intel Xeons. The common thing between all three is the high power CPUs used. That makes them quite different than the Innuos Statement. I don't have a Statement in my system, and I can't really compare.

 

However there are three things I can say:

1. After building several high power servers I can't go back to a low powered server. 

2. The digital source is a complex chain of components (hardware, software, power, etc.) that interact in a super complex way. So it's not about what motherboard you use or what CPU. It's the whole chain that matters - everything from the format of the file you are playing all the way to the output to your DAC. There are so many things to consider in between besides what motherboard and CPU you want to use. What files are you playing? Are you streaming or listening to local files? What software are you using? What operating system? What other processes is the OS running? How do those processes interact with your hardware, software, power supply, mains, etc? Where is the OS stored on? Is it running from RAM? What kind of RAM are you using? What is your local files storage strategy? How is your network configured? I can keep going on and on and list hundreds of these questions. 

That's why a DIY server requires a lot of time for tweaking. If you like doing this, you may be able to tweak a server similar to the builds here to sound better than the Statement. 

On the other hand, companies like Innuos have done all that R&D already and offer a complete solution. They have integrated all the hardware and software for you. You can buy one and enjoy the music rather than doing constant tweaking. 

3. The power supply is the most important part and there isn't much talk about power supplies for these builds (besides the one I showed for the first build, which was only one of multiple power supplies used). You can get a cheap power supply (or two) with one of these builds but you won't be at the level of the power supply on Statement, which has 8 separate rails. Or you can spend a huge amount of money (much more than the computer parts) to get the best power supplies. This would be one of the main determining factors on how a build like these compares to the Statement.

 

So, I guess what I am trying to say is that you can certainly make these builds sound better than a Statement, but that comes with a significant price tag and a lot of invested time. We share a lot of info on this forum, and there are many known DIY recipes that work really well. 

 

2 hours ago, Ianderson said:

Curiously if you were to produce another, what would one of these cost?

I personally have no plans to build computers for other people anymore. It was a good adventure. I learned a lot from these builds. And the feedback has been priceless. Still planning to continue my own experiments and share them, so other people can try them too. And also continue trying what other people have shared - I have learned a lot from the community here! 

Those builds are not cheap and take a lot of time. I can say that I did them for fun and not for money, because if I had to charge for all my time spent, they would be more expensive than comercial servers. 

You can probably have something pretty decent for about $5K. And from there the sky is the limit. I used two of these clocks in one of my builds - they are €1250 each. I also had about $1500-$2000 worth of Mundorf silver/gold wire in that build. Sean Jacobs has a higher spec power supply now than the one he designed for the Innuos Statement - it's £4400 for three rails. The Statement has 8-rails. You do the math. 

 

I am not sure what thought process triggered your question. But if you are wondering if someone else can build you a DIY server that sounds better than the Statement and costs less, I don't really know if that's easily achievable. I feel like you can only pick two items from the list below but not all 3: 

1. Someone else to build it.

2. Cost is no more than the Statement.

3. Sounds better than the Statement. 

And lastly don't forget that a DIY server does not have a good resale value and you don't have support and warranty from a well-established company. 

 

Sorry for the long answer to what seemed like a simple question. 

[Nenon]

I had two Intel 665p M.2 2TB PCIe NVMe cards in an Asus Hyper PCIe card configured as VROC RAID0. They were installed on my Asus Sage dual Xeon build for local music storage. However, since the Asus Hyper card is a full height card, I could not close the top cover of my computer (which is probably fine for now as I am always doing something). 

 

@dminches sent me one of these Supermicro aoc-slg3-2m2 cards to try.

 

It is a low profile PCIe card that can accept 2 x M.2 cards. I moved my NVME M.2 cards from the Asus Hyper to this PCIe card. Booted my computer. But I missed the right time to hit the DEL button to enter the BIOS. Windows booted. And to my surprise everything was immediately recognized, and I was able to access my music stored on those NVME cards right away. The VROC RAID0 worked fine without any changes needed. Pretty cool. My main music folder is just a little bit under 4TB and it fits fine on those two NVME cards. So no immediate need for a new chassis that supports the full height Asus Hyper PCIe cards. I have enough available PCIe slots if I decide to add another 4 TB in the future. 

I haven't compared the sound quality between the Asus Hyper and the Supermicro aoc-slg3-2m2 cards. But given that all they do is split the x16 PCIe slots to multiple x4 PCIe slots, I am guessing there would not be a big difference, if any. The Asus card has a nice big heatsink, but the I/O activity is so little in my case, that I don't think think these NVME cards get hot at all. 

 

@dminches had some problems setting this card on the Asus ROG Crosshair VII Hero. I haven't tested it there yet, but I am glad this is a good replacement of the Asus Hyper card for my dual CPU / Windows server. 

 

[Nenon]

46 minutes ago, Downtheline said:

Nenon, did you need to purchase a vroc key?

 

No. it's not needed for RAID0 when you use Intel NVME storage. 

[Nenon]

11 hours ago, Cazzesman said:

I spent the entire late afternoon and evening reading this thread and the highly informative one started by Neon.    Kudos to you all.

 

My lonely CAPS Lagoon was sadly feeling like the 'hillbilly stepchild' in this field of very heavy hitters.

 

One thing I have not read in any of the threads is costings.    I'm fairly certain us budding DIY PC builders would like to see some prices attached to the various components you have been discussing.  

 

Cases

Power supplies

USB cards

Ram

i9 Motherboards

CPU's,

Heat sinks.

 

They all come at a price along the way.   Please include the RRP of these items where possible. 

 

Whether it be in US $ or the Euro.    I'm presuming these PC's are pushing $10,000 US$.   I'm just trying to get my head around what is possible for the $$$ as a DIY build.

 

Thanks again for all your brilliant posting and ideas.

 

Regards Cazzesman

 

On 6/11/2020 at 8:58 PM, Ianderson said:

Curiously if you were to produce another, what would one of these cost?

 

Since there were a few questions about price, I decided to address those. Let's talk about the cost of these builds. They are not cheap. And there are too many options. It's not easy to cover all of them. But let me try to simplify it a bit. I will be using my recent AMD build for that as I am scared and refuse to calculate the cost of my dual Xeon build (I know it's a lot but I prefer not to know the final number). I'll break it down into 5 categories - computer parts, audiophile cards, clocks and vibration isolation, power supplies, and cables. Please accept my rough numbers. Some of them might be way off, but the intention is to give some idea to the people who have no idea... is it $3K, or $5K, or $10K, or $15K? Let's find out. 

 

1. Computer parts. Here is a sample breakdown. Prices and components change. Let's call it $1500. 

 

2. Audiophile cards. It's $1400 in this case, but it was a lot more when I actually bought them (due to the USD-EUR conversion rates). Let's call this another $1500 for easier math. 

 

3. (Optional) Motherboard OCXO clock and vibration isolation. I rounded up #1 and #2. I am going to round it down this time to balance things out a little and call it $1500 as well just for easier math. 

  

So far we are looking at $4,500 for a top computer. This is with no power supply and stock DC cables. Now comes the difficult part - power supplies.

 

4. Power supplies. That is the part that defines how the computer above would sound. It can easily go from Ok to better to best just by replacing the power supplies and DC cables here.

Let's look at three pricing options - cheaper, more expensive, and cost no object. Please note, I am not saying good, better, best! I am giving products in different price ranges, that's all.

Power supply comparison is a completely different topic outside of the scope of this post, which is focused on pricing. I know some people would ask me regardless, so let me answer this question in advance. I don't know how the DC4 compares to the custom DR or standard SR-7. I am as curious as you are and have been trying to arrange a comparison. But that would most likely take several months. 

 

4a. Cheaper

Get two Keces P8 power supplies. 20V / 8A for the HDPlex and 12V / 8A for the EPS connector. That would be around $1500.

This is not the cheapest option... there are other cheaper options but I would not consider them.

 

4b. More Expensive

Get a 2-rail Paul Hynes Design SR-7 to power the EPS and HDPlex and a JCAT Optimo to power the two JCAT cards. 

That would be around $4,000.

Alternatively, Sean Jacobs can do a 4-rail DC3 LPS.

 

4c. Cost no Object

Get a 4-rail Sean Jacobs DC4. That would be around $6,000. 

 

5. Cables.

The power supplies and the HDPlex have some stock cables included. And you can use those. But better cables are an improvement. Let's look at two options - Ghent Audio (Neotech 7N copper) and Mundorf silver/gold. I am going to make some assumptions for the cable lengths, and that of course may not be the case for everybody, but just to get an idea. Also, in the case of power supply #4b and power supply #4c the stock cables are as good as the Ghent, so 4 out of the 7 cables are not needed if using Ghent. The Mundorf silver/gold are improvement in any case, so I would add all 7. 

 

You need 7 cables.

- power supply to NIC card - 2 feet (16 AWG Neotech / 15.5 AWG Mundorf silver / gold)

- power supply to USB card - 2 feet  (16 AWG Neotech / 15.5 AWG Mundorf silver / gold)

- power supply to chassis (EPS) - 2 feet  (16 AWG Neotech / 15.5 AWG Mundorf silver / gold)

- power supply to chassis (HDPlex) - 2 feet (16 AWG Neotech / 15.5 AWG Mundorf silver / gold)

- chassis to EPS - 1 foot (18AWG)

- chassis to HDplex - 2 feet (18AWG)

- 24-pin ATX cable - 1 foot (18AWG)

 

Here, I explain the cost breakdown of the Mundorf silver/gold cables.

On 6/15/2020 at 12:11 PM, Nenon said:

The Mundorf silver/gold wire, which as the name implies is not a pure silver wire, is by far the best wire I have heard in that application. I would not use this wire for audio interconnects or speaker cables. But for DC cables it is a significant upgrade to any system. This wire is expensive. Here is a cost breakdown for a 3-foot cable:

- 6-feet of wire (what you need for a 3-foot cable) in the US is about $250

https://www.partsconnexion.com/MUNDORF-72180.html#40543

Since Mundorf is a German company, some people in Europe might be able to find it cheaper locally, but those are the US prices. 

- Two Oyaide DC Connectors would be around $50. 

https://www.vhaudio.com/oyaide-dc.html

- Everything else including cotton sleeves, techflex, heatshrink, WBT silver solder, teflon tape, copper braid, etc would be around $25. And if you get those parts from different places and pay multiple shipping fees, the amount grows up pretty quickly. 

- It takes about an hour (maybe a little more) to make one cable, properly dress it, and test it. I guess "your time is free for you" principle applies here. 

Adjust as necessary, add those things up, and do the math. This is not a cheap cable. But if it was sold by one of the big cable makers, it would cost many times the cost if its materials. So, if you think that way, it is actually an incredible value compared to commercial cables. The downside is you have to do it yourself. Or find someone to do it for you. 

 

The reason I am providing a cost breakdown is to make it clear who this cable is for. I would not suggest that people with a $200 LPS start making this cable. I believe it is silly to spend more on the cable than the component. You would be much  better changing the component first. Although, for some reason this cable makes a pretty significant difference even with cheaper power supplies if your system is resolving enough. 

 

To simplify things, I would be using the cost of the wire + $75 per cable ($50 for connectors and $25 for other various materials). 

 

Is it worth spending that much for the Mundorf silver/gold cables?!? In my system - absolutely! 

But also keep in mind that's the cost if you make the Mundorf yourself as DIY. If you have to pay someone else to do it, the cost goes up. 

 

Summary:

One of these DIY computers would cost you anywhere from $4,500 to over $12,000. Here is a breakdown:

 

And that does not include the software. Euphony is an additional $289. 

 

 

[Nenon]

2 hours ago, Cazzesman said:

I have been pondering various upgrades and replacements that could also be reused and future proof to some extent in a new PC build down the track.

 

Based on what I have read (correct me if I am wrong) I would think an improved USB card and LPS to power MB and USB card would be a good step forward.

 

My current set up is

 

System - CAPS (Lagoon) - WS2019 - Audiophile Optimizer v3.0 - Fidelizer 8.4pro -128 gig & 2tb SSD (Music storage) + Sotm USB exp card - Synology NAS DS416 Slim (Backup) - JRiver Ver 25 - USB Curious Cable (Aqvox LNPS each end) into Mutec USB 3+ reclocker into MiniDsp 22D (Dirac Live) via Canare 110 AES into Kii 3 speakers via Ortofon 110 AES, with Kii Controller.  All powered from 2000 VA IsoTran.

 

I currently have a Sotm PCIe exp USB Card in my CAPS Lagoon.    Is the replacement of the Sotm with the JCat XE usb card as easy as swapping one in and out of the MB. (with software tweaks)?

 

Also the Keces 8P looks like a good start.

 

With the USB signal currently being reclocked by the Mutec where do you expect the improvement to come from if the Sotm USB was changed out for the Jcat XE.    Is the Mutec just doing the same thing at the XE?

 

Any thoughts about a slow rebuild whereby the improvements can be carried on into a major rebuild?

 

Regards Cazzesman

 

I am afraid I won't be able to give you a good advice on gradual upgrade path, simply because I don't know some of the components in your system. 

Replacing the SOTM PCIe card with a JCAT USB XE card should be a very simple task. But then you are outputting USB from your computer, converting to AES with the the Mutec DDC, then going through another DDC (MiniDsp 22D), then going to the KII. 

 

What if you replace the SOTM and the Mutec with PinkFaun AES/EBU or SPDIF card? You will lose the ability to play DSD, and your PCM would be limited to 192 KHz, which is probably fine as I am guessing your miniDSP does not support that either. And that is if the Pink Faun card is even supported on your computer. I am not familiar with the CAPS (Lagoon) server and honestly I don't feel like I understand your system enough to be able to suggest gradual upgrades. Sorry, maybe other people can chime in. @bobfa has Kii speakers, but I don't know if he uses Dirac Live. 

There are two other AES options you may want to check (also check if they can fit in your server):

- Lynx AES16e
- RME HDSPe AES

In my system, I prefer the JCAT XE over the SOTM, by a big margin. But on the other hand, when SPDIF was my prefered output, the PinkFaun SPDIF card with ultraOCXO sounded better than any USB card + DDC I have tried. The problem is those PinkFaun cards don't work well with every CPU. 

[Nenon]

1 hour ago, ray-dude said:

Fantastic Nenon!  I did similar back of the envelope math earlier this year, then added in a fudge factor because while I knew I would eventually get to the more premium end of the spectrum, I also knew I would feel compelled to travel the intermediate steps to get there.  That's how I got to spitting distance of jumping on the Extreme train and skipping the intermediate 5 steps.  It's amazing how much power (and to a lesser extent clock) dominates the calculus.  We're basically building power supplies and clock generators with some electronics and storage attached.

You are absolutely right. And the Extreme is a great value when you consider all the parts inside, the time you would save, the support you get, etc.

 

1 hour ago, ray-dude said:

From a DIY perspective, Sean certainly leans heavy into crazy large filters, but even he is an order of magnitude off what Emile is doing.  Has anyone seen another power supply maker take a similar strategy?

I am not so sure about this. I actually think if anything was compromised on the Extreme, it might have been the power supply. But I don't know much about the design, and it would not be fair to make any conclusions. So far I have not seen any compromises on the Extreme, and maybe if I learn more its power supply I would change my mind :). We know Emile is full of surprises!

But I can comment on what I see and if I was to do some educated guesses and speculations, here is what I think. I'll break this down in 3 parts - capacitors bank, transformer(s), and number of rails.

1. Capacitors bank.

What I see is 14 electrolytic Mundorf Mlytic AG capacitors (those are the big capacitance caps) and 3 small value film capacitors (2 Duelund and one Mundorf in a special tankwood stand). Let's focus on the electrolytics as they provide the high capacitance. Those Mundorf Mlytic AG go from 1000 uF to 47000 uF. The most commonly used values in DC LPS are 10,000 uF, 22,000 uF, 33,000 uF, and 47,000 uF. They are typically 30 mm or 35 mm in diameter. I have seen Sean Jacobs using different values in his power supply. He uses four of these per DC rail in his DC4 (and in some DC3's). I don't think he uses the largest values, and I believe this is done for a reason (not related to cost).

A perfect DIY server has 6 rails - 3 rails for the ATX, 1 rail for the EPS, 1 rail for the Network card, and 1 rail for the digital output card (i.e. USB). So, going with 6 x DC4 rails, you will end up with 24 of those Mundorf Mlytic AG capacitors. That's actually a lot more than the Extreme (14). I am not considering the value of the capacitors here. I think Sean is using 10,000 uF by default, but I also think if you need more capacitance he would be able to accommodate it. Whether the end result would be better or worse, I don't really know. I am strictly comparing the capacitance in a vacuum. 

 

2. Transformer(s)

Taiko Extreme uses one 400VA transformer. That's probably not the best transformer for the Extreme. I do believe the transformer is a weak point here.  I power up my dual Xeon server with two 400VA transformers that I believe are of better quality, but I don't want to judge by the cover. Taiko has the transformer geniously installed in a tankwood enclosure - that's brilliant! 

I also always find the transformer to reduce the sound quality when installed in the same chassis as the motherboard. But the Extreme has a thick copper panel between the two. I have tried multiple different panels - they helped but did not completely fix the problem - but I have never done such thick copper panel or the many holes on the Extreme designed to help with EMI/RFI. Let's assume Emile has nailed this down, and that has no negative impact in the Extreme. 

But what if you were to go with 6 rails of DC4? Per Sean's web site, the DC4 is limited to 3 rails: "If you require more than 3 outputs, then you will need to order multiple DC4 units". So, that means you actually need to get two DC4 power supplies for 6 rails. 

That is 2 x 600VA transformers! Again, much (3 times) more than the Extreme. You don't get the tankwood enclosure, but the DC4 has some vibration damping... probably not as sophisticated as on the Extreme though. 

 

3. Number of rails

From our past experiments, we know that the more isolated our server components are with different linear power supply rails, the better. A separate rail for the EPS makes a big difference. A separate rail for the USB makes even bigger difference. And so on. That's why my perfect build has 6 separate DC rails. That has always been better in all my tests. The problem with so many DC rails is heat dissipation. 

I have no idea what Taiko Extreme does for power supply, but I honestly see one rail only. If the Extreme was using many rails with linear regulators, I would expect to see those regulators exposed to the heatsink with adequate space between them, to provide proper cooling as they need to dissipate quite some heat. I don't see anything like that there. If I was to speculate big time, the Extreme is using one rail and the everything is handled by switching mode regulators. Again, Emile is full of surprises, so is it possible that he has discovered something that works better than multiple rails of LPS? Why not... I would not be surprised at all. 

 

All I wanted to say is that although that bank of capacitors looks very impressive, there are DIY options than can give you much more capacitance. But the important part here is the end result. Taiko Extreme has set a new bar and our DIY experiments are not even close at that point. We all need to get back to work :). 

[Nenon]

4 minutes ago, ray-dude said:

The other wrinkle is Emile sharing that high quality clocks actually take away SQ in the Extreme (although he is understandably proprietary about his clock strategy).  He is definitely taking the road less traveled when it comes to power supplies and clocking.  

 

The guy has a lot of experience and knows what he is doing. Here is an example I don't think I have shared...

When I got the PinkFaun ultraOCXO clock for my motherboard, I received a board that looks exactly like the board used on the previous Taiko servers (SGM EVO). Was PinkFaun making the clock boards for Taiko? Or was Taiko making the clock boards for PinkFaun?

Either way, Emile must know a thing or two about clocks! And clocks work hand in hand with vibration treatment and power supplies.