A novel way to massively improve the SQ of computer audio streaming

[Energy]

6 hours ago, ASRMichael said:

Hi, not sure if this is the right place on the forum or not but I seen Romas talking about NUC motherboards. Nuc7i7bnh

 

My TLS db-7 uses a Akasa plato case. Looks really cheap and nasty. I have searched for other options, but nothing I see that can fit. If i was to say, I don't necessary have a budget, but if it was top end would anyone have any recommendations of better pc case?

 

Thanks in advance

Intel NUC7i7DNHE

https://www.quietpc.com/akasa-plato-x7d

 

Good results with Thermal Grizzly Kryonaut.

[Energy]

Thanks for the information @Nenon

 

I was thinking of retrofitting a Pink Faun I²S bridge into an Intel NUC and compare it's sound quality against the USB output. 😊

 

Since the USB output from the Intel NUC is controlled by dPLL, it's clock cannot be upgraded. For this reason I originally wanted to go with an tX-USBUltra to clean up the USB signal before sending it to the DAC, but figured it might be more worthwhile to use an I²S bridge in the NUC for an improved clock output and no longer requiring the DAC's USB input clock to reduce jitter further.

 

I'm not sure if the Pink Faun I²S bridge can be fit into the NUC, but early plans is to use an M.2 to PCI-E x4 adapter and extender cable so it could be side mounted out the back of the NUC. The extender cable will be paid attention to so it can be as short as possible. I'll have it sit above the DAC so the HDMI cable can be kept at 0.3M for optimal performance. Furthermore I have found flat design HDMI cables like WireWorld to sound better than the traditional round ones as it seems the horizontal spacing between the conductors help with crosstalk and ultimately the jitter.

[Energy]

@Nenon

I have a Denafrips Terminator coming in thus why I²S looks very worthwhile right about now. I've sold the Singxer SU-1 I had for a tX-USBultra but would rather use a direct route to bypass all the unnecessary clocks.  

 

So you're saying it won't work on Intel i7 CPU's in general? Even if the unit is an Intel NUC7i7DNHE working as an NAA running AudioLinux? If that IS the case, that would mean it's not a driver issue so it has nothing to do with the operating system and more related to the CMedia CM8888 chip or CPU type as you say. If one does manage to get it to work on perhaps an AMD system, does it have issues upsampling PCM to DSD? Or is the problem more starting with native DSD files itself?

 

And yes, I was going to use a M.2 to PCI-E like the one you suggested and shorten the cable using better quality wires with JSSG360^3 shields. Mylar aluminum foil (15mm) will go spirally down certain data wires to further reduce noise once I can figure out which ones should optimally be grouped up or separated. The teflon and JSSG360 of course would go over them all after completion. 

 

@novaca

Thanks for sharing the HDPlex H1 V3. That case would be a lot better than the Akasa Plato X7D if one chooses to make room for a USB or S/PDIF output card and maybe even fit an LPSU in there. Now I don't have to cut the back of the Akasa just to fit an output card. Great.

[Energy]

9 hours ago, Nenon said:

The PinkFaun I2s bridge is working perfectly fine on my AMD-based system and sounding amazing. However, it can only output PCM. You cannot upsample PCM to DSD with this card. 

 

If this is not a trade secret (since you have a cable company), can you please share some pictures with the DIY community? 

 

It's unfortunate as I prefer using an Intel NUC with a network switch for proper isolation with the added benefit of AudioLinux OS latency.

 

The Windows PC is an upcoming AMD RYZEN 9 3950X powered by HDPLEX 200W LPSU. Enough processing power for HQPLAYER's DSD512 upsampling (poly-sinc-xtr-lp + ASDM7EC) but too noisy to put the Pink Faun I²S Bridge into thus the two box setup.

 

There is likely a jitter increase from using multiple devices (network switch + NUC) and cables (Ethernet + USB) between the PC and the DAC. However I find the isolation to do more for the sound. Perhaps if a person had a PC as quiet as the Pink Faun 2.16x they wouldn't need to do multiple box runs. Given the scenario of a two box setup, as to what you said, the only way to improve from here is remove USB from the chain and do a direct I²S connection to the DAC. The only way for that to work is if the NUC can support an I²S Bridge.

 

Either CMedia CM8888 starts supporting it,  or AMD makes their own NUC, or I build a music server with LPSU that has isolated voltage rails just to implement the Pink Faun I²S Bridge.. To all honesty, I have a feeling a multiple box solution will work better because of the high frequency noise that comes from computer parts (CPU, RAM, ETC) even despite them being cleanly powered. However nothing is certain unless you try it so I'm planning to build both.

 

Every cable or modification I do will be shared with the community to further advancements in computer audio. There is no trade secrets. 😏

 

@Nenon Does the Pink Faun I²S Bridge not support DSD in general? If so that's a deal breaker for me. I listen to nothing but PCM converted to DSD. 😞

[Energy]

I've just purchased the 8GB version of the Apacer Industrial RAM. Will see how it compares to the 8GB version of the HyperX Consumer RAM. CL17 vs C14, but with the difference in industrial grade. Have no heard 2 sticks so I am only able to do a stick vs stick comparison.

 

On other news, it seems like there's now a rise for AMD-based NUC's. The one I saw was the iBox-R1000. I wonder how it sounds compared to an Intel NUC7i7DNHE (Asaka Plato X7D). The cool thing is that they already come in a fanless case and feature displayport with dual Ethernet. If those aren't used, the one interesting thing to point out is their ECC RAM support. TDP is kind of high though at 12-25W but that might depend on how many utilities are being used.

[Energy]

Great find. More information on TruDDR4.

It sounds like according to them, a standard system that takes 1866 MHz has the ability to run at 2133 MHz if using TruDDR4. The system will authenticate the RAM by detecting their unique signature program and thus will load into the profile and run at those speeds. The part I don't understand is where they mention "System x servers and Flex System compute nodes". Since they're RDIMMs and LRDIMMs, it should only be compatible on server motherboards.

 

Slightly different than industrial as those have a lot of other improvements including better BOM to all parts. From what I know is when memory modules are created there are some that are put in certain batches that cannot operate at a higher frequency and so they would get sold at or below that frequency threshold. The TruDDR4 on the other hand has memory modules that succeed the frequency threshold but is sold at that frequency. It is then encoded with a profile so that it can be set to go higher even when the motherboard has a maximum frequency limitation. I could be wrong, but it sounds like the industrial ram would benefit sound quality more given it's fixed bill-of-material (same batch of ram chips), wide temperature, conformal coating, side fill, anti-sulphuration than a ram stick that can go higher clocks.

 

It seems like AMD NUC's aren't going to be released for awhile and even when they do it would be available overseas first. Those markets do better for this type of device I'm guessing. Hopefully those lower power consumption ones can be powered from a Farad Super3. The Epyc 3000 on the other hand looks much bigger than something like a Intel NUC so I wonder if it would still sound as good given more parts/components being used which may result to more noise. The Ryzen 3900 sounds like the perfect CPU to be used in a music server. The TDP puts it just low enough to be used in a fanless chassis.

[Energy]

AMD NUC UPDATE 🤪

 

I will be the first to try an AMD NUC (mini-PC), to put it more precisely the ASROCK iBOX-V1000. <-- Link is provided for purchase.

 

According to Eric Yui who works at MITXPC, he says there is a delivery from ASROCK next week that should come in. Fremont where his company is stationed at is a 30 minutes drive from where I live. I will be picking up the unit once it arrives.

 

 

The unit is similar to a Intel NUC7i7DNHE in the case being a QUAD CORE (V1605B)

But unlike the Intel, it comes with a fanless case so there is no more need for a Akasa Plato X7D Slim.

 

More Information:

ASROCK iBOX-V1000: http://asrock.pc.cdn.bitgravity.com/Download/e-catalog/iBOX-V1000.pdf

AMD Ryzen Embedded V1000 Series (Zen) : https://en.wikichip.org/wiki/amd/ryzen_embedded#V1000_Series_.28Zen.29

 

Comparison:

AMD's V1605B (ASROCK iBOX-V1000)

4C/8T (Threads)

2GHz - 3.6GHz (Turbo)

L2 Cache: 2MB

L3 Cache: 4MB

RAM: DDR-2400

TDP 15W

 

Intel's i7-8650U (Intel NUC7i7DNHE)

4C/8T (Threads)

1.9GHz - 4.2GHz (Turbo)

L2 Cache: 1MB

L3 Cache: 8MB

RAM: DDR-2400

TDP 15W

 

From specification alone, the draw back is that it has a lower turbo clock and L3 cache.

The advantage is having the same TDP, more L2 cache, additional Ethernet input, and support for ECC RAM and Linux.

 

I have good faith that the AMD's CPU performance can beat the Intel's.

From graphics alone their AMD Radeon RX Vega 8 already scores higher than the Intel UHD Graphics 620.
https://laptopmedia.com/comparisons/amd-radeon-rx-vega-8-vs-intel-uhd-graphics-620-the-vega-8-is-nearly-twice-faster-than-its-opponent/

 

Today I spent all morning looking for RAM.

I found these three so far. If anyone can help me that would be appreciated.

THEY MUST FOLLOW THE FOLLOWING: 8GB, 2400MHz, ECC, Industrial or Wide Temperature (WT), SODIMM (260PIN)

1. https://buy.advantech.com/Peripherals-Modules/Industrial-Memory-Solutions-DDR4/model-SQR-SD4I8G2K4SNEEB.htm
2. https://www.priceblaze.com/ts1gsh72v4b-i-Transcend-Memory-Laptop-Memory?ref=gshp&gclid=EAIaIQobChMIq_2tlbGA5QIVaR-tBh1E1wsDEAYYAiABEgLAMfD_BwE
3. https://enoch.systems/en/p/components/memory/m4d0-8gmsqwsj.html?gclid=EAIaIQobChMI7M7Qxq-A5QIVA8ZkCh27OQ43EAkYASABEgLbmfD_BwE

[Energy]

On ldovr.com LT3045 boards the ILIM is connected to GND since monitoring is not used. All the methods on the data sheet also suggest connecting it to GND.

 

If there was a way to cut the output impedance by half I’d do it to all of my LT3045 boards. Please enlighten us. I spent half an hour looking at the data-sheet

[Energy]

I can vouch for Nenon that using silver nano-powder crystals do in fact help, however the price is only justifiable depending on how far you are into your audio setup. If you're happy with the gear you have and want to do some finishing touches, outside of cables, mods, JSSG shielding, this could be a beneficial thing. Personally I prefer using Stabilant 22 as it's cheaper but requires reapplication each time you unplug a contact.

 

Quote

Stabilant 22 is an initially non-conductive block polymer which when used in a thin film between metal contacts becomes conductive under the effect of an electrical field. This occurs at an electric field gradient such that the material will remain non-conductive between adjacent contacts in a multiple pin environment. In addition, Stabilant 22 exhibits surfactant action, as well as lubrication ability providing a single component resident solution to virtually all contact problems.

When applied to electromechanical contacts, Stabilant 22 can provide the connection reliability of a soldered joint without bonding the contact surfaces.

 

More Information: http://www.stabilant.com/appnt15.php

I mix it into a 5mL kit with 100% isopropyl alcohol. The alcohol acts as a carrier agent to disperse evenly onto the contact.

[Energy]

17 hours ago, seeteeyou said:

They do have something else that's as small as it gets

https://www.module-store.de/en/conga-b7e3-amd-epyc-3000/?cat=187

 

I could see myself connecting HDDs to that USB CLEAN port and use it as my quality NAS, not the most powerful hardware for sure but even its Gigabit Ethernet @ 330 Mbps should be no biggie for audio purposes IMHO

 

https://www.allo.com/sparky/usbridge-signature-pcb.html

https://www.allo.com/sparky/usbridge-signature-player.html

 

Great to see more Ryzen / EPYC 3000 options. The 4 core / 4 thread though might not be as good as the current V1605B that comes in the Asrock iBOX-V1000 which does 4 core / 8 thread and with higher clocks too. 

 

USBridge-Signature would make a good endpoint for those who doesn't plan to use a USB regeneration like a tX-USBultra or have the their operating system put on ram boot. I wonder which is better. A less powerful endput with overall lower noise especially on the USB output or a stronger dedicated endpoint running ramboot with ECC/Industrial ram. It's hard to tell. Maybe one day there will be more decent processors with ECC supported RAM and a bunch of ultra low noise voltage regulators being used.

[Energy]

On 11/6/2019 at 1:14 AM, Gavin1977 said:

Probably of interest to others:

 

https://www.udoo.org/udoo-bolt/

 

Embeded Ryzen with ECC support, TDP-down of 12 W and TDP-up of 25 W.  Typical 20V 2.25A (45W) power supply min once memory/SSD is installed, so existing linear power supplies for NUCs like the Paul Hynes SR4 'should' work fine.

 

Also check out the new thread on the LattePanda Alpha - what's really interesting about this Intel Core M3 based system is its 5w/2A amp current draw as this removes the need for costly large amperage linear power supplies to achieve a server/endpoint config.

 

But is the use of Apacer Wide Temperature ECC RAM, as opposed to Non-ECC, worth the extra expense and higher power envelope of the Udoo Bolt V8? Anyone here undertaken comparisions?

 

I ask because the LattePanda Alpha's lower power requirements means that potentially two LattePandas could be used in a server/endpoint config and probably still be powered by a single NUC linear power supply if you're careful about power budget.

 

From what I noticed is that smaller form factors like NUC7i7 have components that act more like a Mini PC works better for transporting audio data than those with extra functionalities (like UDOO BOLT). There are other similar ones available on the market like this: https://www.quietpc.com/tranquilpc-mmdpc-001-v1605b where it is maximized with display ports, although at the cost of using a larger motherboard which would theoretically consume more power and possibly generate more noise.

 

My belief is that the most optimal AMD based endpoint would be one featuring a quad core (like the NUC7i7) with optimization in standard computer functionality while supporting ECC memory. The Asrock iBOX-V1000 accomplishes this for $549 with a fan-less case. Since it's a 4X4 (4" by 4") it should consume less power regardless if other similarly marketed products all follow the 12-25W specification. That wattage depends on whether certain functions like WiFi card, M.2, USB, SSD, Ethernet, or Video is connected.

 

I decided to go with the Asrock iBOX-V1000 method and run mine with AudioLinux via ramboot using a single stick of industrial memory called Innotech 8GB 2400MHz ECC RAM (M4D0-8GMSQWSJ). The device will be powered by a Farad Super 3. If there is room inside the unit I will add a JCAT USB Card and connect it to the M.2 socket using a converter cable. This should help isolate the noise that is generated by the endpoint and keep it away from the USB controller chip, thus reducing jitter and noise at the output before it is handed to a SOtM tX-USBultra. The tX-USBultra cleans up the USB signal even more before passing it onto the DAC. Also the tX-USBultra will send three of it's available clock outputs from it's internal sCLK-EX clock synthesizer to the endpoint so it can replace the clocks with higher phase noise (System, Ethernet Input, and USB output). I will post a photo once I finalize it sometime this month.

 

Glad everyone is enjoying Industrial RAM and ECC. I have found that it sounds a little more 3D when used in both the server and endpoint. Sound is a little denser with more air. Transients slightly easier to hear, but barely. Smoother transitions.

 

I am currently looking for a better storage device for my music server. It is using a Samsung 960 PRO M.2 and recently I found out that some M.2 NVME's have a built-in ECC called LDPC (Low Density Parity Check) to correct both hard-bit and soft-bit information during decoding. It is said that this lengthens the lifespan of the chips and so we could assume that ECC RAM will last longer too. Either way, upon doing more research I slightly concluded that running an SSD with ECC could lead to better sound quality since it could be powered by a linear power supply's 5V. Doing so may improve how these chips perform. Given the lower noise, it may help with timing and errors.

[Energy]

I have a NUC7i7DNHE to compare to. Since the Intel NUC doesn't support ECC, I'll use a non-ECC HyperX CL14 when comparing the two endpoints. This way sound quality during A/B test is all endpoint related and not linked to RAM quality.

 

Asrock iBOX-V1000 supports ECC since it's APU (V1605B) processor support it within it's memory controller.

https://en.wikichip.org/wiki/amd/ryzen_embedded/v1605b

 

 

UDOO BOLT V8 has the same APU (V1605B) as the Asrock iBOX-V1000.

 

Many of these small form factors Mini PC's won't have a PCI-E slot since the cases are too slim to allow a PCI-E card to be vertically mounted. If a case had a space for horizontal mounting, a PCI-E to PCI-E riser cable is required for the liftoff distance. So there is no difference in using a riser cable or a M.2 KEY M (2242/2260/2280) to PCI-E converter cable, since a cable is still needed. M.2. and PCI-E operate at PCI-E X4 so they are compatible.

 

 

In order to accommodate the JCAT USB Card, I will have to CNC an opening for the USB output and DC input, then drill two holes for a bolt and nut to hold the internal bracket in place. It should be fairly simple yet still a tedious process.

 

The real question is whether using a cable will rid some of the benefits of having the card since the length of the cable now becomes susceptible to EMI/RFI which when added together with length can add jitter. Proper actions should be taken such as reducing the length of the cable by cutting and re-soldering, shielding wires that have voltage running through them so they don't affect the data wires that are running parallel to them, and using a shield that goes around the whole cable to further protect the data wires. Additional methods can be added for so called "better" performance such as using UP-OCC solid core or stranded LITZ wire instead of the ribbon cable that the converter cable comes with, heat-shrinking each wire in neoprene heatshrink for low tribolectric effect, adding space between conductor and shielding to reduce capacitance and electrical absorption (neoprene does this), using copper foil instead of aluminum foil for better shielding properties, and applying Stabilant 22 to all contact points before pushing them into their respected slots.

[Energy]

Let's say one decided to get the Apacer SSD with "Built-in hardware ECC". With operating read/write speeds of over 500MB/s, it makes you wonder if going any higher would be beneficial for audio purposes. If speeds matter, the reading speed should be more important as that is mainly what is being used to validate the digital bits. Another thing to think about is whether the built-in hardware ECC in Apacer SSD's is any different than the ECC Engine used for Low-Density Parity-Check (LDPC) used in their M2 lineup. 

 

Having pointed out those things, which would sound better?

1. Apacer SSD (Built-in hardware ECC) with 520MB/s read speed

2. Apacer M2 (LDPC ECC) with 2660MB/s read speed

 

Furthermore, what if someone powered the SSD with an external linear power supply's 5V?

Powering the M2 on the other hand with a 3.3V seems less likely as its attached to the motherboard.

 

Seeing that power consumption on the Apacer M2 goes as high as 1,725 mA, for those barely meeting power requirements from maybe running an AMD 3950X DSD upsampling server, this wouldn't be very ideal.

[Energy]

26 minutes ago, dminches said:

 

Is there a sound advantage to using a PCI card for the M2 as opposed to putting it in the M2 slot on the motherboard?

 

 

PCI card to M2 risers make for a longer signal path and relies on two contact points of insertion.

M2 directly connected to the motherboard should sound better unless the chip that controls the PCI-E Gen 3/4 was put closer to the PCI-E card slots and doesn't have to go through the various PCB layers and noise upon reaching the M2 slot, but given multi-layer PCB motherboard designs and proper routing this shouldn't pose a problem.

 

At this point a riser card could be comparable to running an SSD over SATA cable, but in a lesser degree. It's more favorable to use a PCI-E to M2 because the distance isn't nearly as long as with making SATA connections (data/power). But one thing to note that can influence sound quality between the two is the power going into the PCI-E lanes loop through the motherboard (potentially picking up noise) opposed to power coming directly from the ATX PSU and into the SATA drive.

[Energy]

https://www.chiphell.com/thread-2170925-1-1.html
 

pretty insane

[Energy]

I wanted to give it a shot here on my end but it was a hard pillow to swallow without AudioLinux as a renderer. I wouldn't be able to use HQPLAYER for upsampling.

 

Thinking of purchasing it anyway to try next month with JPLAY FEMTO so I'll report back then.

[Energy]

Nice find Nenon.

 

For those wondering, It's an optical USB cable with optional USB 2.0 Micro-B input for 5V Power.

The optical part adds jitter but the jitter must be minuscule compared to what is gained from blocking out server noise.

 

It would be interesting if the cable was shorter with Type A and Type B male for a direct connection from server to DAC. I can modify the connector ends, but the long run still bothers me as it wouldn't be a direct comparison to an audiophile USB cable (since it's shorter). However, if the idea is that it makes whatever USB cable that is used with it even better, then it should be better no matter the length.

[Energy]

16 hours ago, [email protected] said:

Looks nice

Yes,  someone already tried.... 

 

Are those additional electrolytic's on the input and output required for the stack? I thought tantalum's were better for the output (lower ESL), better transient response.

[Energy]

55 minutes ago, kelvinwsy said:

U should try what I have done! JSSG your power cords with a shield grounding lead and then tie this to the Earth pin in your power bar/power conditioner. Bet you will hear the background noise reduce by another 2-3 dba

 

I concur. However I found it to perform better when an independent conductor was used for ground, Lead and Neutral twisted to improve on magnetically induced voltage coupling, and JSSG to be floating with at least 2 separated layers that are tied at the ends for the noise currents to rotate into infinity.