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

[lmitche]

Perhaps remove the Nvidia card and try again?

[lmitche]

Funny, I get 27 process after AO installation, ultimate mode, remove drivers .... and shell replacement to HQplayer.

 

Later I disable three services msdtc, DHCP client, and Base filtering engine.

[austinpop]

Funny, I get 27 process after AO installation, ultimate mode, remove drivers .... and shell replacement to HQplayer.

 

Later I disable three services msdtc, DHCP client, and Base filtering engine.

 

That's cuz you're a god, Larry, while we are mere mortals.

[austinpop]

Perhaps remove the Nvidia card and try again?

 

No can do - I'm just using a vanilla Dell XPS 8700 desktop, and the only video ports (VGA and DVI) are on the Nvidia card. Without it, I'd be headless.

[romaz]

The thread:http://www.computeraudiophile.com/f22-networking-networked-audio-and-streaming/network-card-clock-upgrade-31824/

It was pretty hot there☺. There were things pretty technical in that thread.

Thanks for the link, GL. Obviously, a lot of contentious debate about the wisdom of replacing the clock on a LAN card and some good points were brought up by some. I have my own speculative comments which I cannot definitively prove but they are borne from consistent observations.

 

The point was brought up repeatedly that if you disconnect the ethernet cable, music will continue to play. This is not really a new revelation and I have posted on it before. Roon themselves have indicated that they employ a fixed memory buffer where if you disconnect the incoming data stream, music will continue to play. In my testing, this buffer lasts for about 1-2 tracks and some players have even larger buffers where whole albums can be stored and played back from memory. The implication here is that the quality of the ethernet cable should have absolutely no impact on SQ and neither should the clock from the preceding LAN port since playback is occurring off of a memory buffer within the server and not from a steady ethernet stream.

 

Here is another way to look at the concept of buffers (ie memory playback). As far as I am aware, without exception, anything with a clock also has a buffer that is associated with it because for a clock to do its job properly, it requires a continuous and uninterrupted stream of data which is achieved by means of a buffer. Taking this concept further, all interfaces that contain clocks would, therefore, also contain buffers and this would include any network device such as NICs, switches and FMCs and of course digital components like music servers and DACs. In the same way that your music server buffers the ethernet stream it receives before rendering and reclocking it, your USB DAC also buffers the incoming USB stream and so all such devices, to some degree, employ "memory playback."

 

Now, looking specifically at DACS, if the argument that any cables or clocks that precede its buffer should have zero impact on SQ was true, then that would mean any jitter introduced by a bad clock and RF noise introduced by a bad cable should have zero effect. Conversely, any jitter that was reduced by a good clock or RF noise that was reduced by a well shielded cable should also have zero effect. It would, therefore, be logical to deduce that devices like the mR or sMS-200 should have zero impact on the DAC with respect to SQ and that as long as the stream is "bit-perfect", all sources should sound the same, whether it be a Raspberry Pi or an Aurender W20. As we know, this is a common argument of many "bits are bits" theorists and yet, for many of us who are willing to listen and compare, our ears have told us differently and that feeding a component a signal with high integrity can make a difference.

 

As to the impact of clocking (or reclocking), I think that anything that has direct contact with the signal has the potential to imprint on that signal (for better or worse) and that this imprint can potentially be carried forward all the way to your speakers. Those that have done comparative testing between SSDs and compact flash cards know that there are notable SQ differences between these media despite the understanding that the signal is ultimately buffered by the OS into memory downstream. As I have given the subject of clocking more thought, I can think of few things with greater potential impact on SQ than a clock. Not only does a bad clock directly contribute to jitter but a poorly powered clock is also a source of substrate noise given its location within the signal path and given how many bad clocks are in the typical signal path, my guess is that their collective impact is not insubstantial. Whether replacing a certain clock (like a LAN clock) will make a difference, I think it depends on where in the chain it will be placed and whether it will be followed by a bad clock. If placed before a stock NUC or Mac Mini, it probably would have minimal impact. If placed within the "direct connection" path and followed by something like the sMS-200 Ultra, based on my observations thus far, I would expect it to have a tremendous impact.

 

Any pricing info for replacing the regulator on the Intel NIC?

 

$50 per regulator and $20 per capacitor.

[shoom]

‘’If placed within the "direct connection" path and followed by something like the sMS-200 Ultra’’

Yes that’s exactly what I have in mind but with the standard SMS200.

I have an un-used Netgear switch for a testing mule just to see what improving the clock may do (if anything). I’ve built a few DAC’s in my time and the clock scheme looks pretty much the same. I have a TentLab shunt (and quite a few other quality PSU) to try powering it with.

Clock on order.

Soldering Iron warming.

Ears open and expectations realistically low.:-)

 

 

[greenleo]

Thanks for the link, GL. Obviously, a lot of contentious debate about the wisdom of replacing the clock on a LAN card and some good points were brought up by some. I have my own speculative comments which I cannot definitively prove but they are borne from consistent observations.

 

The point was brought up repeatedly that if you disconnect the ethernet cable, music will continue to play. This is not really a new revelation and I have posted on it before. Roon themselves have indicated that they employ a fixed memory buffer where if you disconnect the incoming data stream, music will continue to play. In my testing, this buffer lasts for about 1-2 tracks and some players have even larger buffers where whole albums can be stored and played back from memory. The implication here is that the quality of the ethernet cable should have absolutely no impact on SQ and neither should the clock from the preceding LAN port since playback is occurring off of a memory buffer within the server and not from a steady ethernet stream.

 

Here is another way to look at the concept of buffers (ie memory playback). As far as I am aware, without exception, anything with a clock also has a buffer that is associated with it because for a clock to do its job properly, it requires a continuous and uninterrupted stream of data which is achieved by means of a buffer. Taking this concept further, all interfaces that contain clocks would, therefore, also contain buffers and this would include any network device such as NICs, switches and FMCs and of course digital components like music servers and DACs. In the same way that your music server buffers the ethernet stream it receives before rendering and reclocking it, your USB DAC also buffers the incoming USB stream and so all such devices, to some degree, employ "memory playback."

 

Now, looking specifically at DACS, if the argument that any cables or clocks that precede its buffer should have zero impact on SQ was true, then that would mean any jitter introduced by a bad clock and RF noise introduced by a bad cable should have zero effect. Conversely, any jitter that was reduced by a good clock or RF noise that was reduced by a well shielded cable should also have zero effect. It would, therefore, be logical to deduce that devices like the mR or sMS-200 should have zero impact on the DAC with respect to SQ and that as long as the stream is "bit-perfect", all sources should sound the same, whether it be a Raspberry Pi or an Aurender W20. As we know, this is a common argument of many "bits are bits" theorists and yet, for many of us who are willing to listen and compare, our ears have told us differently and that feeding a component a signal with high integrity can make a difference.

 

As to the impact of clocking (or reclocking), I think that anything that has direct contact with the signal has the potential to imprint on that signal (for better or worse) and that this imprint can potentially be carried forward all the way to your speakers. Those that have done comparative testing between SSDs and compact flash cards know that there are notable SQ differences between these media despite the understanding that the signal is ultimately buffered by the OS into memory downstream. As I have given the subject of clocking more thought, I can think of few things with greater potential impact on SQ than a clock. Not only does a bad clock directly contribute to jitter but a poorly powered clock is also a source of substrate noise given its location within the signal path and given how many bad clocks are in the typical signal path, my guess is that their collective impact is not insubstantial. Whether replacing a certain clock (like a LAN clock) will make a difference, I think it depends on where in the chain it will be placed and whether it will be followed by a bad clock. If placed before a stock NUC or Mac Mini, it probably would have minimal impact. If placed within the "direct connection" path and followed by something like the sMS-200 Ultra, based on my observations thus far, I would expect it to have a tremendous impact.

 

 

 

$50 per regulator and $20 per capacitor.

 

Thank you Romaz for your reply and I'm with you. I left that thread because some of the discussions were too high level and some are too difficult for me.

I believe that everything in the path matters and we're trying to find the bottlenecks and remove them (then the 2nd order becomes 1st order bottleneck) until diminishing return is encountered.

 

Given that the NIC is connected to the net-player (mR, sms200,...) directly, this would be my 1st order bottleneck ☺. I guess an Audio grade NIC will come out soon given your great discovery of the direct connection and I truly look forward to it.

 

How did you ask Sotm to change the regulator and the regulator? Is it an open service?

[Nouchka]

Hi,

 

I have disabled 20 services after AO in ultimate mode on Win 2016 GUI, I have now 21 services running.

I don't know why these services were still running after AO script applied...

 

list of services:

windows firewall

network list service

network location awareness

windows event log

windows time

windows connection manager

quality windows audio video experience

ike and authipsec keying modules

function discovery resource publication

distributed transaction coordinator

windows audio

windows audio endpoint builder

network setup service

wlan autoconfig

tcp/ip netbios helper

computer browser

workstation

AMD external events utility

 

Funny, I get 27 process after AO installation, ultimate mode, remove drivers .... and shell replacement to HQplayer.

 

Later I disable three services msdtc, DHCP client, and Base filtering engine.

[lmitche]

That's cuz you're a god, Larry, while we are mere mortals.

Anyway, I still pare everything back manually at every reboot to 21 processes with HQplayer/roon.

No can do - I'm just using a vanilla Dell XPS 8700 desktop, and the only video ports (VGA and DVI) are on the Nvidia card. Without it, I'd be headless.

Headless is good, just try tightvnc. Sure there is no video output on your motherboard?

 

You may be very surprised at the sq improvement without a monitor and video card.

 

Nevertheless the machine should boot without the video card. Tightvnc has been very reliable here.

[ted_b]

Larry,

do you not use HQPlayer? I found CUDA offloading to be invaluable with things like multichannel upsampling.

[Evo1668]

Well today I finally received my Klimax DS/2 streamer upgraded to Katalyst aka Klimax DS/3......Just substitute any positive expletive, for what an expertly implemented clocking/Dac architecture can do for the sound, just unbelievable! And I would not of thought I could improve on the Linns Ethernet connection directly into a switch.....how totally wrong I was!

So my final source setup is

JS2 powering 2012 MacMini with MMK

ethernet in to the MacMini, with two bridged Ethernet connections available for:-

LPS1 powering MicroRendu (UPNP bridge mode)

Klimax Katalyst DS

 

Holy Cow!!

[austinpop]

Since it's Linn, you must mean Holy Kow!

 

 

Sent from my iPhone using Computer Audiophile

[lmitche]

Larry,

do you not use HQPlayer? I found CUDA offloading to be invaluable with things like multichannel upsampling.

Ted, yes I use HQplayer 2 channel and upsample everything to dsd512. With a 6700k I have no use for a Nvidia card.

[ted_b]

Ted, yes I use HQplayer 2 channel and upsample everything to dsd512. With a 6700k I have no use for a Nvidia card.

 

Ah true, for 2 channel.

 

Question: can my NVidia card be bypassed for monitor use (in case I need to do bios stuff I'd plug into mobo connector) and still be active for HQP Cuda support? Guess this is a better question for Jussi, but you are up there in my mind.

[Superdad]

I've been using a display emulator with my Mac Mini just to allow quicker screen sharing connections. Apparently it does this by turning on the GPU. I kinda wondered it this would be detrimental to SQ. I'll pull it and see.

 

This is the device by the way:

https://www.amazon.com/dp/B00FLZXGJ6/ref=cm_sw_r_cp_tai_arnWybJ3D0SKZ

[/url]

 

I do not agree (at least for Mac mini) that it is a foregone conclusion that eliminating the monitor (or monitor EDID emulator dongle) will sound better. Without an attached monitor or EDID dongle, the CPU performs the GUI tasks. With a monitor or dongle, those tasks get put off to the GPU.

 

So in some cases it might sound better to keep the GPU active and the load off the CPU. I have not made that comparison in a very long time. My system is much more sensitive/revealing these days, so maybe I'll try it again.

[kennyb123]

I do not agree (at least for Mac mini) that it is a foregone conclusion that eliminating the monitor (or monitor EDID emulator dongle) will sound better. Without an attached monitor or EDID dongle, the CPU performs the GUI tasks. With a monitor or dongle, those tasks get put off to the GPU.

 

So in some cases it might sound better to keep the GPU active and the load off the CPU. I have not made that comparison in a very long time. My system is much more sensitive/revealing these days, so maybe I'll try it again.

 

I'll be interested to hear what you discover. I pulled the emulator dongle and didn't notice any change whatsoever in SQ. Since I so infrequently use screen sharing I just left it off.

 

You are right that the CPU would have to perform the GUI tasks without monitor or dongle. But if neither a monitor nor a dongle are attached, does the CPU actually get burdened with GUI tasks?

[seeteeyou]

Took me a fairly long time to go through 1,000+ posts, let's see what I was able to find then.

 

Regarding that Mac Mini / Windows / SDXC combo, someone was able to run Windows 10 off micro SDXC but Parallels 9 was used

 

https://discussions.apple.com/message/29687534#message29687534

 

And then someone created an *.VHDX image of Windows 10 while booting that off USB flash drive

 

https://forums.macrumors.com/threads/bootcamp-windows-10-from-external-drive.1815672/page-7#post-23038526

 

Maybe that VHD could also work for SDXC cards after following these instructions?

 

Deploying Win 7 to Boot From VHD with MDT 2010

https://blogs.technet.microsoft.com/deploymentguys/2009/03/03/deploying-win-7-to-boot-from-vhd-with-mdt-2010/

 

Guide: Deploying Windows 7 to a Macbook as a VHD

https://social.technet.microsoft.com/Forums/en-US/9ef457f2-5152-4c1a-a02e-a1b717e1ed24/guide-deploying-windows-7-to-a-macbook-as-a-vhd

 

Yet another alternative could be Grub4Dos

 

USB Boot Grub4dos into .VHD

http://reboot.pro/topic/19604-usb-boot-grub4dos-into-vhd/

 

---

FYI - for many audiophiles in Asia, their favorite way to boost SQ turned out to be "heavily" optimized (i.e. very lightweight) Windows 10 residing in RAM

 

http://www.myav.com.tw/bbs/showthread.php?threadid=20478104

 

Link to that optimized VHD

 

http://uptobox.com/lqj0vl80qsq6

 

We've got a very similar thread here as well

 

http://www.computeraudiophile.com/f10-music-servers/running-win-os-purely-ram-27402/

[HeeBroG]

I agree that a Mac Mini is nice setup for a music server. It's housed in an attractive and fairly resonant-free aluminum chassis. Although Apple had different intentions for doing it, in shielding their RAM and CPU in a stainless steel cage, they also effectively provided a nice RF shield against 2 components that generate quite a bit of it. It also has very short signal paths and putting the SD card reader on a PCIE bus and allowing the OS to boot from it was a genius move. Despite its CPU fan, this thing generates only 11dB of noise and is essentially silent. Obviously, it lends itself to being powered by a 12V source and your MMK is the best way I know to do it. On top of that, their hardware just works which is the benefit of their closed system -- all their components have been vetted for compatibility and reliability. Compared against a NUC, I find it to be a better starting platform and many used Mac Minis can be found cheaply on

Hi Romaz,

 

Do you think the soon to be released SoTM sps-500 could power an MMK Macmini?

It's rated at 12V/5amps.

 

G

[Theobetley]

Hi Romaz,

 

Do you think the soon to be released SoTM sps-500 could power an MMK Macmini?

It's rated at 12V/5amps.

 

G

What is the SoTM sps-500? What function does it perform?

[HeeBroG]

What is the SoTM sps-500? What function does it perform?

 

It's the new battery power supply from SoTM that may be more cost effective than a Uptone JS-2/Paul Hynes SR-7 LPS.

[Theobetley]

It's the new battery power supply from SoTM that may be more cost effective than a Uptone JS-2/Paul Hynes SR-7 LPS.

Thanks

[romaz]

Took me a fairly long time to go through 1,000+ posts, let's see what I was able to find then.

 

Regarding that Mac Mini / Windows / SDXC combo, someone was able to run Windows 10 off micro SDXC but Parallels 9 was used

 

https://discussions.apple.com/message/29687534#message29687534

 

And then someone created an *.VHDX image of Windows 10 while booting that off USB flash drive

 

https://forums.macrumors.com/threads/bootcamp-windows-10-from-external-drive.1815672/page-7#post-23038526

 

Maybe that VHD could also work for SDXC cards after following these instructions?

 

Deploying Win 7 to Boot From VHD with MDT 2010

https://blogs.technet.microsoft.com/deploymentguys/2009/03/03/deploying-win-7-to-boot-from-vhd-with-mdt-2010/

 

Guide: Deploying Windows 7 to a Macbook as a VHD

https://social.technet.microsoft.com/Forums/en-US/9ef457f2-5152-4c1a-a02e-a1b717e1ed24/guide-deploying-windows-7-to-a-macbook-as-a-vhd

 

Yet another alternative could be Grub4Dos

 

USB Boot Grub4dos into .VHD

USB Boot Grub4dos into .VHD - Grub4dos - reboot.pro

 

---

FYI - for many audiophiles in Asia, their favorite way to boost SQ turned out to be "heavily" optimized (i.e. very lightweight) Windows 10 residing in RAM

 

Win10 14986RamOSÄÀ¥X - MyAVµøÅ¥°Ó±¡ºô

 

Link to that optimized VHD

 

Download 14986 rar

 

We've got a very similar thread here as well

 

http://www.computeraudiophile.com/f10-music-servers/running-win-os-purely-ram-27402/

Thank you, seeteeyou! You have provided me different options to ponder. Running the OS in RAM sounds like a great thing to do and I have read through the threads that you provided. It seems to accomplish what Daphile accomplishes although Daphile seems simpler to implement and allows for a bridged LAN configuration. Once my build is completed, it is my intention to try Daphile and compare. Have you had a chance to compare the two?

 

I have some other questions for you:

 

1. Have you (or has anyone) had the opportunity to try and compare Windows Server 2012R2 (or Windows Server 2016) + AO (in Core mode or minimal server mode) off of a compact flash drive (or SD card) against Phasure's version of optimized Windows 10 in RAM?

 

While I have not personally had the privilege of experiencing W2012R2 + AO from a CF card, here are my personal observations in my system thus far:

 

A. Using Roon with either W2012R2 + AO in minimal server mode vs W2012R2 + AO in GUI mode vs Win 10 Pro + AO (only GUI mode is possible) vs standard Win 10 Pro + Fidelizer Pro vs MacOS (El Capitan) + CAD script, W2012R2 in minimal server mode sounds best and removal of the GUI with AO seems to result in a definite improvement in SQ. I know that Windows Server 2016 in Core mode will sound even better and I expect (based on Phil Hobi's comments) that Roon will be possible in this configuration at some point. I realize there are other "non AO" methods of optimizing Windows 10 Pro but I have further found AO's proprietary sound signature and digital filter options to allow for sound tuning that benefits both the sMS-200 and mR (especially the mR) without impacting the bit-perfect stream. As I have stated, with AO, I have found these sound signatures and filters alone to be worth the price of AO.

 

I realize there are also optimized versions of Linux to consider including Daphile, Euphony, SonicOrbiter and SOtM's own Eunhasu and it's quite possible some of these options could be tuned to be a bit better than the best that Windows has to offer but thus far, I have not found anyone to publicly state that, especially against Windows Server 2016 in Core mode (without taking Roon into consideration). I do also concern myself with compatibility. First of all, not all of these Linux options allow for Roon and the option to bridge your LAN ports. Moreover, other programs like Spotify and Pandora have now announced the option for lossless streaming and who knows what other options will become available in the near future but almost certainly, all of these options will be available on Windows.

 

B. In my comparison of running the OS off an SD card vs PCIE SSD (NVMe mode) vs SATA SSD vs SATA 2.5 inch HD, there is a clear and obvious preference for running it off an SD card and I suspect running it off an SLC compact flash card will be even better. Of course, this then begs the question of whether running an OS from RAM would sound better than running the same OS from an SLC CF card and so this is what I personally would like to know. Looking at the links that you provided and trying to read through the enormous Phasure forum, I don't recall anyone specifically providing a comparison between the two.

 

On face value, it would seem that running an OS entirely from RAM should sound better but from Phasure's own forum, with even the most optimized version of Windows 10 Pro, it suggests you would need 16G of RAM (8G wouldn't be enough). This presents its own challenges because RAM creates its own noise. As we know, RAM's voltage rating is provided with every stick of RAM we buy. With DDR3, it's 1.5V, with DDR3L, it's 1.35V, with DDR4, it's 1.2V, and eventually, with DDR4L, it is expected to be 1.05V. While this is great from the standpoint of "running cooler", it says nothing about current draw which, I believe, is much more important when considering how much noise something will be potentially injecting back into the ground plane. For example, a 1V source that draws 12A would have considerably greater negative impact on SQ than a 12V source that draws 1A even though both components consume the same 12 watts.

 

Since RAM manufacturers don't generally list current draw in their spec sheets, I used the online OuterVision Power Supply calculator (OuterVision Power Supply Calculator) which provides estimates of how much current under load a 4GB module of RAM will draw against a 16GB module, how much DDR4 will draw against DDR3, and how much 2 sticks of RAM will draw compared to 1 stick of RAM and here is what I found. To go from 4GB of DDR3 (which is what I currently use and what I expect to use with my new build) to 16GB of DDR3, if this calculator is to be believed, under load with 16G of RAM, you would draw a whopping 4A more current compared against 4G of RAM. DDR4 fares a bit better and the draw is 3.3A more when you jump from 4G to 16G but this is still huge. As for 1 vs 2 sticks of RAM, using 2 sticks of 8G RAM draws no more current when compared against just a single stick of 16G RAM.

 

In the best case scenario with DDR4 where you draw 3.3A more with 16G vs 4G of RAM, this is considerably more current draw than what you would get from running an OS off a compact flash card, especially when you consider that Windows is supposed to be a "write heavy" OS. You also have to consider that with a compact flash card as an OS drive, you can externally power it. Of course, there are other things to consider that may weigh in favor of RAM including the impact of the flash drive controller, the much higher latency of compact flash cards and the impact of the SATA cable and so this is why I would love to have someone who has experienced and carefully compared both types of setups to comment.

 

2. With Phasure's method of running Windows 10 in RAM, do you know if Roon is possible? Based on what I read on Phasure's forum from posts in mid-2016, this was not possible. It isn't possible with Daphile either. While this isn't a deal breaker for some, it is for me. While SQ is paramount, I have come to also value the other qualities that Roon brings to the table. I have nearly 5k albums in my library but thousands more Tidal albums and all are conveniently catalogued by Roon and all of these albums are easily searched, skinned and queued. My overall experience with Roon is far and above anything that JRiver, A+, iTunes, Foobar, MPaD, Squeezebox, HQP, BHE, etc. have been able to offer and while Roon is definitely a personal preference, my overall listening engagement is considerably higher with it. In the past, I struggled with this guilty pleasure because it was at the compromise of SQ but since Roon 1.2 and with extremely lightweight, very low impedance and low noise Roon Ready endpoints like the mR or sMS-200 Ultra, this compromise has disappeared and my very best SQ today is with Roon.

[romaz]

Hi Romaz,

 

Do you think the soon to be released SoTM sps-500 could power an MMK Macmini?

It's rated at 12V/5amps.

 

G

SOtM hasn't listed the specs for the sPS-500. It states it is a hybrid PSU and so I don't yet know what that means but my guess would be that it involves batteries or supercaps in some fashion. If this is the case, it would be unlikely to be able to power a Mac Mini. Even Vinnie Rossi's Mini Pure DC-4-EVR supply can't power a Mac Mini.

 

EDIT: I stand corrected! You are right. I just noticed there are now specs listed and indeed, 12V/5A is listed as an option which is quite surprising considering how small this PSU is. I am definitely interested to know how well this performs.

 

What I will say is that my experience with PSUs with switchable voltages is that they come with challenges. With Paul Hynes' PSUs, to do this requires cascading regulators (which is a good thing in terms of added line noise rejection) but as you go from 12V down to 7V, for example, to maintain 5A of output results in more heat generation that could result in stability issues. Exactly how the sPS-500 will address this will be interesting to see.

[beautiful music]

I'm a new comer here in this thread.In fact i succeed to make a static bridge between my microRendu direct connection to my laptop using LAN cable and wifi.

 

But i don't know what is the right configuration in jRiver MC21 to run my stored music in my internal HD.

 

Any advice is highly appreciated.

I Bought also sMS-200 but i'm trying to see microRendu performance before to listen to sMS-200

[beautiful music]

Also i faced a problem as my internet connection is interrupted despite i specified a static IP to my wifi network.

 

Any thoughts