DIY Project High Performance Audio PC with high quality wiring

When I build my last audio PC, I went through some learning phases. For example, that there is not a single PC cable manufacturer in Europe that offers cables from an audiophile point of view. Instead, cheap stripes of the worst kind are mostly used. The cables are often too long and are laid crisscross or rolled up in the PC. It was only with ghentaudio from China that I was able to achieve satisfactory cable quality.

 

 

While fooling around, Bernd von fis Audio and I came up with the idea of building a high-performance audio PC with high-quality cabling together. The signal paths should be short. The audio PC should offer enough power to drive even complex filters or DSD EC modulators of the HQPlayer. It should be completely fanless, i.e. passively cooled. A clean linear power supply is mandatory. The case should be protect against resonances.

 

With this thread I want to report on our project progress. We are grateful for tips.

 

1. Shopping List

 

Below I list the components and the reasons that led to the selection. Of course we try not to let our project budget get out of hand. But we also want to install the best parts or at least very good components in the audio PC.

 

1.1 HDPLEX 2nd Gen H5 Fanless Chassis

 

This is a comparatively inexpensive and elegant case, which offers space for a motherboard with an ATX form factor. With 43cm width it has the "usual" HiFi format and with 11cm height it should fit in anywhere. The case is available in silver (our model) and in black. The passive cooling looks very high quality. It is intended for cooling CPU's with a power (TDP) of 95W.

 

1.2 ASUS ROG MAXIMUS XI FORMULA

 

Choosing the mainboard was not easy. In any case, it should be a gaming board since the power supply is usually better. It almost became the MSI MEG Z390 ACE because it has very good reviews. Unfortunately, it has no HDMI connection, so a graphics card should have been installed. For certain settings, such as BIOS optimizations, a screen has to be connected. So the choice fell on the expensive ASUS ROG MAXIMUS XI FORMULA board. It has some features that are appealing to a passive system. Namely, a very good cooling of important components. It is built very robustly and has an Aquantia 5G LAN connection, which invites you to experiment. Just take a look.

 

1.3 Intel Core i9-9900K (95W TDP), 3.6GHz - 5.0GHz OctaCore - HD Graphics 630

 

The CPU is well known to me for its stability and smooth operation. The new flagship i9-9900KS would also have been nice, but since the 5GHz clocking across all cores is not possible with a passive-cooled system, the i9-9900K remained. With this CPU, conversions from PCM to DSD 512 or the use of EC modulators for DSD 256 are no problem at all with the HQPlayer. 

 

1.4 Corsair DOMINATOR® PLATINUM 16 GB (2 x 8 GB) DDR4 DRAM 2400 MHz C10 memory kit

 

Due to their robustness and low latencies, these 16GB RAM are also well-known.

 

1.5 Samsung 970 EVO Plus MZ-V7S250BW - 250 GB SSD

 

Two fast M.2 sockets can be used in the mainboard. A 250 GB SSD is sufficient if the music comes from NAS or from a streaming provider. If you want to have the music in the same device, you can connect 2 x 2 TB = 4 TB. If that is still not enough, connect further SSDs to the 6 x SATA 6Gb / s ports. There is enough space in the HDPLEX fanless chassis.

 

1.6 JCAT USB Card Femto

 

Experience shows that a dedicated USB card with an external power supply is worthwhile. At least when the DAC is controlled via USB. Other cards are also conceivable as an option. For example from Pink Faun.

 

1.7 JCAT NET Card Femto

 

With this card, extremely slim and interference-free network connections can be configured under Windows.

 

1.8 HDPLEX 400W DC-ATX

 

An ATX mainboard also requires an ATX power supply. The HDPLEX 400W DC-ATX converter has proven itself very well here.

 

1.9 Keces P8 Single 20V / 8A + optional 12V / 8A

 

The HDPLEX 200W Linear Power Supply (LPS), which is known to have dissolved in smoke, has not proven itself. Since then, I've been extremely concerned about security. The Keces P8 series has
 
 Over Voltage Protection (OVP).
 Over Current Protection (OCP).
 Under Voltage Protection (UVP).
 Over Temperature Protection (OTP)

 

Reading the power consumption is very practical. The LPS are quite handy, flawlessly processed and load stable. With normal use they only become lukewarm. We will also connect a 12V / 8A Keces directly to the 4-pin EATX 12 V Power Connector. This is supposed to bring something good in terms of sound.

 

The DC power sockets still to be selected on the HDPLEX fanless chassis should, however, be flexible in use so that any power supplier can be connected.

 

1.10 PC and DC cables

 

There will be some surprises. Bernd will draw from his broad range of cables. The plug connections are just as important. Gold-plated Molex pin connectors are already set for the 24PIN / 8PIN / 6PIN / 4PIN-Molex connections. Another playground are the DC cables for the external power supply.

 

1.11 Minimization of resonances

 

Sensitive electronic devices particularly benefit from stable, resonance-free device feet. We will experiment with different BF Magic Spacers from fis Audio. Balls made of glass, precious metal or other materials are placed on the discs and deflect vibrations.
 

Stay be tuned. 😉

 

Gabriel

Let me be the first one to comment. Happy to see more builds documented, and this one looks promising already. Please see my comments below.

 

8 hours ago, StreamFidelity said:

1.4 Corsair DOMINATOR® PLATINUM 16 GB (2 x 8 GB) DDR4 DRAM 2400 MHz C10 memory kit

 

Due to their robustness and low latencies, these 16GB RAM are also well-known.

Have you considered using Apacer RAM? I think you will like it.

I am pretty much done trying consumer memory. No matter what speed or latency they have, the Apacer always sounded better in my system. My latest discovery was that you can use ECC RAM on those LGA 1151 motherboards. It works as non-ECC RAM but sounds a little better. 

 

8 hours ago, StreamFidelity said:

1.6 JCAT USB Card Femto

 

Experience shows that a dedicated USB card with an external power supply is worthwhile. At least when the DAC is controlled via USB. Other cards are also conceivable as an option. For example from Pink Faun.

You may want to hold off to see Marcin's new reference USB card. 

 

8 hours ago, StreamFidelity said:

1.8 HDPLEX 400W DC-ATX

 

An ATX mainboard also requires an ATX power supply. The HDPLEX 400W DC-ATX converter has proven itself very well here.

If your budget allows, try the HDPLEX 800W DC-ATX. It fits nicely in the chassis with the ASUS ROG motherboard:

But you will have to drill holes on the bottom of the case. And yes, I will have a super short ATX cable, that's why it's lined up like this . 

 

8 hours ago, StreamFidelity said:

The DC power sockets still to be selected on the HDPLEX fanless chassis should, however, be flexible in use so that any power supplier can be connected.

I ordered some Jaeger connectors and a thick carbon fiber plate. Planning to replace the stock small power connector plate that comes with the case with a custom carbon fiber plate and two Jaeger connectors on it. I will post more about it when I make it. I think it will look really good. And those connectors are top notch. 

 

8 hours ago, StreamFidelity said:

1.11 Minimization of resonances

 

Sensitive electronic devices particularly benefit from stable, resonance-free device feet. We will experiment with different BF Magic Spacers from fis Audio. Balls made of glass, precious metal or other materials are placed on the discs and deflect vibrations.

This is interesting. I may ping you on PM for more details. 

FYI - here's the latest (and greatest) 2666MHz UDIMM that's recommended by Marcin

 

D31.23185S.001

https://www.mouser.com/ProductDetail/Apacer/D3123185S001?qs=Cb2nCFKsA8prqNusSnCNZw==

 

75.B93H0.G000B (2400MHz UDIMM) should be his previous favorite.

1 hour ago, dminches said:

 

Did Larry tell you this?  He owes me one since the one I bought 6 weeks ago was defective.

 

I was checking something on his web site yesterday and saw them showing as in stock. 

 

1 hour ago, auricgoldfinger said:

Marcin found this 2400MHz UDIMM to sound the best.

 

http://jplay.eu/forum/index.php?/topic/2786-ram-adventures-extremely-interesting-results/page-2#entry49239

I see other people commented on this already. Marcin's post was about the 2666MHz RAM. That was the "new" RAM he referred to. 

 

29 minutes ago, dminches said:

What is the downside or lost opportunity in using ECC ram if the mobo uses it like non-ECC ram?  If one is using ECC ram shouldn’t they try to find a mobo that supports ECC?

I don't really know but the ECC RAM sounded better as non-ECC on the last build I posted about. Perhaps it's not because of the ECC bit but the quality of the memory. I compared a 2666MHz Non-ECC Apacer with the same version of ECC 2666MHz Apacer running as non-ECC. I liked the ECC RAM running as non-ECC better. 

 

There is another member on this forum performing the exact same tests (2666MHz Non-ECC Apacer vs. 2666MHz ECC Apacer running as non-ECC) but may take him some time to report. 

7 hours ago, pranabindu said:

Wow, what a great thread!  I'm kinda new to DIY audio PC builds, so I apologize if my questions seem dumb.

 

I got stumped trying to pick a motherboard.  What audio performance characteristics should I be looking for?  Chipset stats?  BIOS tweakability?  Once you pick a CPU, there are SO MANY motherboard options, it is overwhelming.

 

About the video (to be able to connect a monitor, to set up what will be a headless PC): if the motherboard has a video output, does the CPU have to be rated for video, too?

 

There is mention here of RAM "quality."  To what does that refer?  The latency numbers?  Any specific stats I should look out for?

 

Does the PCI-E speed (8x, 4x, etc.) make a difference in sound quality if my plan is to use the Pink Faun HDMI card?

 

Pink Faun seems to put its OCXO's all over the motherboard.  They are expensive.  Is there a less expensive solution for a stock motherboard?

 

Thanks for any advice you might have.

 

It depends what kind of computer you are building (eg. one box or two box setup).

 

If you are building a headless computer, the device should still have a HDMI or Display Port for you to initially set up headless ramboot via Flash Drive with the pre-loaded config. It's been found that a NUC (nano unit of computing) with a decent performing CPU that is powered by a low noise linear supply sound better than some less powerful devices, however this requires a power supply that can handle at least 15W which is a little over 2A of current.

 

In order to reduce power consumption further one can disable certain processes from being used in BIOS such as certain audio controllers, WiFi cards, etc. Some NUC's come with a heatsink fan to cool down the CPU but this induces noise and consumes more power so it is advised to put the circuit (motherboard) into a passively cooled case instead, preferably with good performing thermal paste like Thermal Grizzly Kryonaut for the CPU DIE.

 

For awhile many people have been fond of the Intel NUC7i7. The device comes with two problems. The first problem is that Intel NUC's do not support ECC ram. The second problem is that their DC input takes a wide range of voltages hence there is more noisy DC to DC conversion taking place. The device has to convert the higher voltage (normally 19V) to lower voltages like 3.3V, 5V, and 12V for it's components. Outside of noise, this voltage drop also generates more heat. 

 

On the discussion of RAM/MEMORY, If you are creating a headless NUC with an operating system like AudioLinux, that said operating system has to first be loaded onto a flash drive with the essential configuration required for it to perform as an endpoint (NAA) over the network. Both the flash drive and the RAM needs to be 4GB to support said operating system. Some operating systems like Euphony may require more. It is said that using two sticks in dual channel helps to increase bandwidth and lower the latency of CPU/RAM interaction (data queue's) but doing so involves using another stick that operates at 1.2V. This means more power consumption. Some people have found that one stick to sound better but this can mainly be due to the device not being fed with adequate power thus it performs better with just one stick. The NUC's take something called SODIMM RAM. They are DDR4 and come in many different frequencies and timing (CL). It's preferable to use the one with the highest frequency supported by your NUC's motherboard but what people have found to be the best is called ECC Industrial RAM from the brand APACER. It may not be rated with the highest operational frequency, but people have found that Industrial, ECC, or both combined to sound better than higher frequency or low CL timing RAM. Industrial specified memory modules basically RAM with integrated circuit (IC's) chips from the same batch so their operating voltages are more aligned with one another. ECC (Error-Correcting Code) is a type of register IC placed on the RAM to correct the soft and hard bit errors that computers sometime tends to make. This means that without ECC your outputted files aren't truly "bit perfect" until they are corrected. Intel NUC's in correlation to ECC is that they do not support it. Only their workstation/server computer motherboards do. For both of these problems combined I decided to go with an AMD NUC instead called ASROCK IBOX-V1000. The two advantages is that it receives a set voltage of 12V at the DC input and supports ECC enabled RAM. Both of which has lead to better sound quality after many A/B tests.

 

To address your question on video. CPU's deal with processing tasks while an APU (CPU + GPU) does the same but has video embedded on it. Most NUC's are APU's or has a CPU with a separate GPU on it so most that you purchase that is recommended here will come with either HDMI or Display Port 1.4 on it.

 

Now the NUC can be used in two different ways. Either as a server or as an endpoint. As an endpoint in order to reduce further noise on the circuit it is better to run it headless and operating system loaded onto the RAM via RAMBOOT. If you use it on a server, although it works, even with 4 CPU core it does not have the processing power as a full fledged computer would have when processing music that require heavy filters and modulators. For example if you were to upsample PCM music to DSD, you will be limited to how good a filter/modulator you can use due to CPU processing power limitations. This is one of the reasons why there is an uprising of audiophiles wishing to build passively cooled computer music servers that have more processing power. The problem with this is there are more variables that go into building a computer of this caliber. First off, since the TDP (thermal design power) of a larger computer (eg. Full ATX) requires more power, one cannot easily power it with a lower current linear power supply like with the NUC. It also generates more heat so in order to passively cool it you would need something like an HDPLEX H5 case. In terms of RAM you too can use ECC/Industrial for the larger server. But instead of SODIMM it is called UDIMM. UDIMM is unbuffered memory which means the CPU's memory controller sends tasks directly to UDIMM's DRAM bank of memory modules. RDIMM is better in the case that most of them are also ECC but have an extra chip to register given workloads. This chip organizes the data and prioritizes it in order before sending it to DRAM. The benefit here is it takes the workload from the CPU's memory controller. It is very beneficial for server's with multiple CPU's as they require it, but still somewhat beneficial for a single CPU. The unfortunate part is that most RDIMM's are not made to Industrial specification. However since they are made in smaller batches, they are more likely to perform in closer tolerances than standard unbuffered DIMM's. 

 

Another benefit of using a larger server with a larger motherboard (like an ATX) is that there are PCI-E and PCI available for a plug in card such as the one you mentioned (Pink Faun I²S). It is common to use cards like JCAT Net Card or USB CARD FEMTO to isolate the Ethernet or USB controller chip further away from the noisy motherboard circuitry and power the card with cleaner external power (5V/1.5A). This improves the measured jitter of the output device since the controller circuit is now better isolated, operate under clean power, and utilizes a lower phase noise clock (Crytek CCHD-957). These cards can be installed in PCI-E x1, x4, x8 or x16 lanes. The lanes with the higher multiplier have more bandwidth and thus performs faster. If the card only has enough pins for a particular multiplier (eg. PCI-E x4), using it on a lane with higher bandwidth won't necessarily improve it. You'll however want to connect it to the PCI-E lane that makes a direction connection to the CPU rather than the one that routes to the chipset first.There are cheaper cards out there like the Matrix Audio Element-H (USB) that comes with an aluminum shielded cover. It is hopeful that other brands will one day do the same for extra EMI/RFI rejection.

 

Clocks/Oscillators with lower phase noise can be used to replace the lesser performing ones on located on the motherboard or even on the add-on cards. Each aftermarket clock should come with it's own circuit and powered with clean power. Pink Faun does have this in the form of their OCXO modules. The higher priced OCXO clocks usually leads to lower phase noise plots, but do note that their power input is equally important alongside how well they are decoupled/dampened and temperature controlled (using an oven). In order to replace stock clocks you will need to know how to solder or know someone who can with some technical know-how in circuitry. A cheap option is to purchase over the counter clocks like the Crystek CCHD-575 in the frequency you want to replace from websites like Mouser or Digikey and use an adapter board for the clock to sit on with wires leading up to the location you want to replace. It is advised to power said clock with a low noise linear voltage regulator like an Analog Devices LT3045 rather than using the noisy voltage supplied by the motherboard. These regulators can be purchased from LDOVR.com but required to be connected to the 5V of a power supply so it can drop the voltage down to 3.3V for the clock. A slight higher priced option is to buy pre-made clock modules, however I advise you look at it's phase noise before you purchase them. Most of them will do better than the stocks but if you don't properly pay attention to the phase noise, maybe not by much. A higher priced option would be mounting a SOtM sCLK-EX clock module to an Akasa 2.5" SSD/HDD bracket by drilling holes to mount the PCB board by using 5-10mm hex standoffs and imperial #4-40 or M3 machine screws. There should be another drilled hole made to fit a GX16-2 DC input connector and a two wires leading from the connector to the clock module's power input (9V/1A). The clock module has four clock outputs that can be programmed for the frequency that is required. It stock it has a lower phase noise than Crystek and even lower when a Masterclock is connected to the masterclock input of the device. What occurs when the connection is made is that it takes the incredibly low phase noise 10MHz signal of the Masterclock and uses it as a reference point for it's clock circuit to generate/output even lower phase noise.The even more pricey ones are from Pink Faun such as their "Ultra OXCO". The downfall of that is that they only work for one location. I personally use the sCLK-EX which only occupies one mounting bracket in the back of the case but can tap into four different locations. Normally you'll want to connect the clocks to SYSTEM, CHIPSET, OUTPUT (I²S, USB, ETHERNET), and whatever clock that controls the display you're using (HDMI or DISPLAY PORT). If you are using both JCAT NET and USB card then don't clock the display and clock both of these cards instead.

 

In conclusion, having an ATX music server will be more ideal in being able to use the top of the line digital filters such as the ones available on HQPLAYER thanks to the higher CPU processing power of the server. Aside from processing power it allows you use more isolated add-on cards for lower jitter performance as well as allowing more space to be used for aftermarket clocks. The only inconveniences is that the CPU cannot have too much CPU cores as this raises the temperature beyond the passive cooled chassis's allowed TDP and in order to power this kind of server with a linear power supply you will need something with like a lower end HDPLEX 200W or higher end Paul Hynes SR7. This sole reason is why people on the forums right now are trying out various linear power supplies in the voltages needed by the motherboard (3.3V, 5V, 12V) in different locations (EPS and ATX). One could simply use a DC to ATX converter but these create over 10mV (10,000uV) of noise which is way beyond a typical audiophile linear power supply that normally are single digit uV. Thus it is recommended to disable all unused IC's in BIOS to further save on power consumption. In general, the music server has lots of DC to DC converters on board despite being fed clean power so it is advised to used a network switch (eg. EtherRegen) to galvanically isolate computer noise from reaching the DAC (since Ethernet is inherently galvanically isolated). With this setup it requires a NUC to be placed after the network switch and before the DAC to receive music files from the server and send it to the DAC. This method has less noise since the NUC is uses much less current and is fed a linear power supply. It doesn't do much heavy level processing on it's own as it's job is to just simply hand over the audio files. This is the reason why having a set input voltage for the NUC is better than having one with a wide input (12-19VDC) as those can create more noise that can eventually irrigate to the DAC. This is the reason why USB reconditioning devices like UpTone ISOREGEN or SOtM tX-USBultra help with sound quality when placed after the USB output of the server or NUC. Lastly, one can improve the clocks within the NUC also, but with small form factors and passive cooling there is very little space. 

 

There is two ways I could think of for having an end game system.

1. Taiko Audio SGM Extreme (One Box Setup)

2. Custom Built Server & Endpoint Based Audio Chain (Multiple Box Setup)

 

Not sure which would be better between the two..

45 minutes ago, StreamFidelity said:

@Energy 

Thank you for this very good contribution. It's like a compendium. 👌

 

And even this really ingeniously new designed Taiko Audio SGM Extreme has a disadvantage: to my knowledge it will not be able to render the EC modulators in HQPlayer with DSD 256. This requires a clock frequency of at least 4 GHz over several cores. It's ok, because the target picture is different. Our goal is to render with EC modulators.

 

Look: whatsbestforum

 

 

That's what I was going for. 😉

Now if anyone asks questions just link it to that. It'll make life a little simpler.

 

I concur. We both find DSD upsampling superior to anything else available. I have tried JPLAY Femto but found HQPLAYER far better. Even when using streams like Tidal, although JPLAY sounds good, using Roon as a front end that's connected to Tidal and feeding audio streams to HQPLAYER to have it upsampled from PCM to DSD is still better by a long shot. How organic the sound is, timbre, decay, openness in sound stage/air, and especially the layering of music or how it's projected is much more ear candy.

 

Right now I am using an AMD 3950X. Even when overclocked all 16 cores still fail to get poly-sinc-xtr-lp to play ASDMEC7 at DSD512. While everyone else "fauning" over the Taiko SGM Extreme, we are looking at passively cooled 8GHz CPU's to handle DSD512 instead. 😅 I definitely miss the sound stage of DSD512 so much that I sometimes go back to the AMSDM7 512+fs modulator just to play DSD512 with acoustic music for a better sense of head space.

 

Back to the server build discussion.. I have a few things to note

1. Apacer Industrial RAM should sound better than any consumer ram regardless of CL timing

2. Intel Optane for the operating system should sound better due to it's lower latency

3. Thermal Grizzly Kyronaut thermal paste will be better for the heatsink

9 hours ago, StreamFidelity said:

 

1.11 Minimization of resonances

 

BF Magic Spacers from fis Audio @Nenon I'll get more information, PM will follow.

 

I've had tremendous results using custom made polished hardened Aluminum bowls with high precision ball bearings.  By far the biggest impact on my DAC, but very audible benefits on the server as well.  Interesting to see a commercial option.  After some experimentation, I ended up settling on a shallow bowl (5" diameter).  The Magic Spacers look like they have a much deeper bowl.

 

Looking forward to hearing your findings and following along this project!