Jump to content
IGNORED

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


Message added by The Computer Audiophile

Important and useful information about this thread

Posting guidelines

History and index of useful posts

Most important: please realize this thread is about bleeding edge experimentation and discovery. No one has The Answer™. If you are not into tweaking, just know that you can have a musically satisfying system without doing any of the nutty things we do here.

Recommended Posts

6 hours ago, Johnseye said:

Why use the NUC as an endpoint when the sMS-200ultra or the ultraRendu provide the same functionality?  Was it because there wasn’t one of those devices to compare against the NUC?  Are the hard drives on those devices noisier than the eMMC on the NUC?  The sMS and uR can be powered by an LPS-1.2 or SR7/4 with less power.

 

Oops John, no hard drives in ultraRendu or sMS-200ultra, though the latter allows for hanging a USB drive off the extra port (yuck).  OS is loaded in RAM upon boot up--at least with the Rendus, I have no firsthand experience with the SOtM boxes.

Link to comment
On 8/8/2018 at 4:18 PM, austinpop said:

I believe the key to Roy's findings may be related to the use of an SR-7 DR rail, and the use of eMMC.

This is the key I think (along with the modded/reclocked switch of course).

The NUCs with EMMC card may well be what I have been waiting for. For the past 6 months or so I have been using a simple and cheap SD card player which only has one clock - a very simple and cheap system for I2S out with stunning results - thanks to the TirNaHifi thread. 

http://www.tirnahifi.org/forum/viewtopic.php?t=4541

 

Not so convenient as the music files have to be loaded on cards but once you get used to it and cleanly powered, without question the best sounding source I have heard by a long shot. This is the main reason I have not been tempted to try more expensive paths. 

 

These NUCs running on a card however could provide all the SQ benefits of a card player with access to Roon and Tidal streaming as well. Worth a look as Roon Rock could probably run on an 32gb card. I think I'm going to have to give it a try, so I'm adding this to my list of experiments.

 

What great thread that I keep coming back to! Many thanks Rajiv for continued curating and sharing.

Topaz 2.5Kva Isolation Transformer > EtherRegen switch powered by Paul Hynes SR4 LPS >MacBook Pro 2013 > EC Designs PowerDac SX > TNT UBYTE-2 Speaker cables > Omega Super Alnico Monitors > 2x Rel T Zero Subwoofers. 

Link to comment
2 hours ago, auricgoldfinger said:

Yes, that's the one. I bought the half and three quarter inch braids. Crazy expensive, so I didn't buy much. Early signs are that it improves SQ once again. Details in the next day or two.

Pareto Audio aka nuckleheadaudio

Link to comment
1 hour ago, Superdad said:

 

Oops John, no hard drives in ultraRendu or sMS-200ultra, though the latter allows for hanging a USB drive off the extra port (yuck).  OS is loaded in RAM upon boot up--at least with the Rendus, I have no firsthand experience with the SOtM boxes.

 

So they are nearly identical to the NUC Roy is talking about.  Slightly different hardware, but they both appear to load the OS into RAM.  The OS is different, but not sure how much that has an impact.

Link to comment
5 hours ago, lmitche said:

John, you have the tls switch, will you be experimenting?

 

 

I will but not sure what path to take.  I think there are some limitations on the frequencies the sMS-200ultra and ultraRendu can pass.  I could be wrong, but that may be why Roy went to this NUC as a Roon bridge instead of one of the others as Roon ready.  Maybe @Superdad can clarify whether the uR can pass PCM 384, 768 or DSD512.  Do Roon ready devices use the Roon bridge?

 

I have my NAS going through the switch now.  I can bridge it directly to the PC and see how it compares.

Link to comment
10 minutes ago, Johnseye said:

The OS is different, but not sure how much that has an impact.

Roy is running Dream OS from thelinearsolution.com.  It's supposed to be tuned for low latency.  The name alone doesn't convince you? 

Pareto Audio AMD 7700 Server --> Berkeley Alpha USB --> Jeff Rowland Aeris --> Jeff Rowland 625 S2 --> Focal Utopia 3 Diablos with 2 x Focal Electra SW 1000 BE subs

 

i7-6700K/Windows 10  --> EVGA Nu Audio Card --> Focal CMS50's 

Link to comment
50 minutes ago, Johnseye said:

 

I will but not sure what path to take.  I think there are some limitations on the frequencies the sMS-200ultra and ultraRendu can pass.  I could be wrong, but that may be why Roy went to this NUC as a Roon bridge instead of one of the others as Roon ready.  Maybe @Superdad can clarify whether the uR can pass PCM 384, 768 or DSD512.  Do Roon ready devices use the Roon bridge?

 

I have my NAS going through the switch now.  I can bridge it directly to the PC and see how it compares.

Ok so I have a theory, not fully baked yet, but I'll share it anyway.

 

The idea is inspired by the mumetal experiment done today. The braid was installed on the JSSG 360 lush cable and the 2 foot jssg 360 Gotham power lead from the lps-1.2 to the Iso Regen. Once again a significant increase in SQ was delivered. Just more of everything, detail, timber extension, bass...you name it. I am likely moving over a 100 amps of dc current to upsample in hqplayer. Emi/rfi radiation must be relatively high.

 

Roy said he had a long Ethernet cable between the server and nuc renderer. He also said it sounds best with the switch near the renderer. 

 

Ok here goes: The distance between the server and renderer means the emi/rfi from the server can't impact the sound of the renderer. As such the only source of emi/rfi is from the renderer itself or the tls switch or either or both of the power supplies. Given the low current in play here, I'm guessing less then 2 amps at 12 volts,  any emi/rfi signal strength is much lower then the server emi/rfi emissions.

 

Do you have a long Ethernet cable?

 

Larry

 

P.S. this would also explain why music sourced from Tidal and my local hard disk have the same sound quality.

 

 

Pareto Audio aka nuckleheadaudio

Link to comment
1 hour ago, lmitche said:

Ok so I have a theory, not fully baked yet, but I'll share it anyway.

 

The idea is inspired by the mumetal experiment done today. The braid was installed on the JSSG 360 lush cable and the 2 foot jssg 360 Gotham power lead from the lps-1.2 to the Iso Regen. Once again a significant increase in SQ was delivered. Just more of everything, detail, timber extension, bass...you name it. I am likely moving over a 100 amps of dc current to upsample in hqplayer. Emi/rfi radiation must be relatively high.

 

Roy said he had a long Ethernet cable between the server and nuc renderer. He also said it sounds best with the switch near the renderer. 

 

Ok here goes: The distance between the server and renderer means the emi/rfi from the server can't impact the sound of the renderer. As such the only source of emi/rfi is from the renderer itself or the tls switch or either or both of the power supplies. Given the low current in play here, I'm guessing less then 2 amps at 12 volts,  any emi/rfi signal strength is much lower then the server emi/rfi emissions.

 

Do you have a long Ethernet cable?

 

Larry

 

P.S. this would also explain why music sourced from Tidal and my local hard disk have the same sound quality.

 

 

 

The ethernet cable between the switch and PC is long.  Probably about 20 ft.  Between the NAS and switch is about 1 ft.

Link to comment
4 hours ago, lmitche said:

..

The idea is inspired by the mumetal experiment done today. The braid was installed on the JSSG 360 lush cable and the 2 foot jssg 360 Gotham power lead from the lps-1.2 to the Iso Regen. Once again a significant increase in SQ was delivered. Just more of everything, detail, timber extension, bass...you name it. I am likely moving over a 100 amps of dc current to upsample in hqplayer. Emi/rfi radiation must be relatively high.

 

...

 

 

In my MuMetal conduit (solid conduit, not braid) experiments on USB, I am finding that it affects the sound enormously adding unbelievable details and tightening bass. Unfortunately it compresses the soundstage and makes the brighter sound further away and less organic. After listening for longer periods of time, I find that without MuMetal conduits is better. Now working on shorter conduits that only cover part of the USB cable.

 

I have the conduits over my 12V PoE DC cables supplying my router but they don't seem to make that significant a difference but I still have to confirm this again when I find some time.

Link to comment
On 8/8/2018 at 8:12 AM, ElviaCaprice said:

Nice to see you get around to a NUC.  Now if that NUC was reclocked with a sCLK EX for three of it's clocks and you added a PCIe 1X lane like I did with an adapter ( I can't tell if there is a  mSATA on that mobo), you could add a tXUSBexp PCIe card for the last clock.  

Good to hear of your success with NUC, SR7 powering it (as I do), interesting on the OS from memory.  But I've found little in SQ change regardless how much I strip my Windows 10 OS down driven from an external 5TB 2.5" HDD powered by an LPS-1 and connected to the NUC via SATA II.  I see a single SATA connector on that mobo of yours.  The Jetway NUC can also hold 8GB of memory.  But I use 4GB.

 

http://www.jetwaycomputer.com/NU93S.html

 

Thanks for the feedback, you do all the hard work running around in circles for SQ on finding the heavenly server stream.

 

 

No doubt that this NUC board can be further improved, Mark, but it is impressive to hear how good this thing sounds "as is."  And running an OS from memory doesn't guarantee low latency if the OS is inefficient and has lots of unnecessary processes/threads running and if inefficient drivers are being utilized.  Sometimes, the bottleneck is in software and not in hardware which is why it was interesting to see the wide range of 10-150X less latency that was reported when this OS was run from RAM vs a PCIe NVMe SSD.  Also, RAM should have way more than just 10-150X less latency than an NVMe SSD and so obviously, a well-tuned OS matters.

Link to comment

It seems rather ironic that the Direct Bridge method can be improved by inserting a switch into the direct path.

Seeing as the main purpose of these new super switches is to improve SQ, rather than act as a switch, then what's to stop the designer putting that technology into the input end of their renderer devices?

 

So, if we take the upcoming Uptone switch, which is built from the ground up, why can't the key SQ ingredients go into a new superultraRendu? And save on an extra box (with p/s and cables etc) - for those that don't need any extra switching capabilities?

 

I know it's different companies in this particular example, but the principle stands.

Link to comment

If booting off eMMC (without storage devices via USB / SATA / PCIe etc.) into RAM were THAT good, we could only speculate how that's gonna sound if we're going completely diskless by means of either iSCSI or PXE boot

 

http://www.ccboot.com/lan-boot.htm

 

BTW, it would be mighty interesting to compare NUC with something even simpler @ 300 bucks

 

https://up-shop.org/up-boards/97-up-squared-pentium-quad-core-8gb-memory64gb-emmc.html

 

Their latest model $289 UP Core Plus should be even simpler than its predecessor

 

https://up-shop.org/up-ai-edge/231-up-core-plus.html#/63-up_core_plus_sku-atom_x7_e3950_8gb_ram_64gb_emmc_on_board

 

$49 carrier board is coming in September

 

https://up-shop.org/up-core/202-up-core-carrier-board-high-speed.html

Link to comment
On 8/8/2018 at 8:18 AM, austinpop said:

 

In light of Roy's findings, Adrian had sent me a generic Intel NUC box, based on the same mobo Roy mentioned, with his Dream OS. One key difference was his NUC had an SSD, not the eMMC. In my tests, I powered the NUC with my SR-4, as I did not have access to the SR-7, nor the JS-2, at the time. My results did not mirror Roy's findings wrt the superior dynamics he reported. My SE was still outperforming this NUC in all respects.

 

I believe the key to Roy's findings may be related to the use of an SR-7 DR rail, and the use of eMMC.

 

I should point out that Adrian's upcoming DS-1 streamer is an embodiment of this same NUC design, in a productized form. Additionally, Adrian will offer an OCXO module, that can be configured to drive one of: the system, USB, or Ethernet clocks. Only one, not all 3. This is the dilemma with fixed OCXOs in a streamer - where do you apply the low phase-noise clock.

 

I have asked Adrian to send me a sample of his DS-1 as soon as he can so I can evaluate and provide feedback. Specially now that I have access to Eric's SR-7, and will soon (a relative term!) have my own SR-7. I've asked him to provision it with OCXO on the system clock, based on the rationale that I have an OCXO switch upstream, and a Ref-10-clocked USB regenerator (tX-USBultra) downstream.

 

I'll keep y'all posted.

 

Regarding power supplies for this NUC board, yes, it scales extremely well to power supplies and better than my DFI board.  This board can accept 12-19V.  During my testing with my SR7s, 19V sounds a little better than 12V but DR (double regulated) 12V sounds considerably better than 19V.  The ideal SR7 for this board would likely be a DR 19V which I do not have.

 

There is a language barrier when communicating with Adrian as English is not his first language and so he did not fully grasp the significance of the eMMC drive with this board that I selected.  The board he ultimately chose to go with in his upcoming DS-1 streamer is an even better board than what I have and it was the board I had wished to use originally.  This board is the Intel NUC7CJYB which also incorporates a 32GB eMMC drive but unfortunately, it cannot be purchased standalone through standard channels.  You would have to purchase the much more expensive NUC7CJYSAL kit that includes the board along with Windows 10 and a heavily plasticized NUC chassis.  You would have to explant the board from the stock Intel chassis and preferentially install it into a fanless chassis although because this board is so new, I have not yet found a company who is making an alternate chassis for this board.  I would suggest against using the stock Intel chassis since it uses a fan that is powered by the motherboard that undoubtedly will be adding it's own electrical noise to the ground plane. 

 

The significance of this other board is that it incorporates the newer Celeron J4005 processor that has an identical 10w TDP but provides a larger size cache of 2MB per core (as opposed to only 1MB per core with the board I have).  In theory, this should result in even lower latency although whether this results in any real world improvement in SQ is not something I have knowledge of.  I believe Intel is supplying TLS with these boards on an OEM basis and that they have also made arrangements with another company to build a very attractive and solidly made fanless aluminum chassis for their upcoming streamer.  As Rajiv mentioned, there will be the option to outfit this streamer with a single OCXO.  Of course, it will come with their Dream OS which will run from RAM and can be powered by any suitable 12-19V CPU of your choosing (since my unit consumes up to 15w during bootup, I don't think an LPS-1.2 will work).  I don't believe they've finalized their pricing yet but for those who would prefer a turnkey solution that comes with support, this could be a very good option.

 

FYI, I do not have any financial arrangement with TLS nor do I plan to.  They have not provided me anything for free and thus far, I have not purchased any of their products.  In exchange for the use of their OS, I have provided them feedback and that is all.

Link to comment
15 hours ago, Johnseye said:

I’ve been thinking through the outcomes of this post for the past two days.  Here’s my takeaways and resulting questions.

 

1.       The OS drive has played a significant impact in his tests.  Mac mini with Mac OS on SD card, Mac mini with Windows OS on NVMe SSD, drive of the Zenith SE, TLS Linux OS on RAM, NUC with TLS OS on eMMC.  Of all these options the NUC with TLS Linux OS on the eMMC drive that sounded the best to Roy.  My results with the Optane also blend well with these findings.

2.       The Celeron processor on that NUC could not upsample DSD, but if there is a similar board with i7 proc this could be possible.  I suspect an i5 or i7 will still require a fan.

3.       I suspect this is early in Roy’s testing, but this board could be modified by SOtM to utilize their sCLK-EX.  This would also allow the use of a REF10 for the board.

4.       The bridge connection with endpoint does bring us back full circle.  Many of us have or still do this.  Question is what the NUC brings to the table over another endpoint.

5.       Roy found little difference in the upstream source, comparing his Mac mini with the Zenith SE, especially with the modified clock switch in between it and the endpoint.  I my own listening experiments with Rajiv we both could hear a difference when the server was a noisy PC vs a customized sCLK, low power PC.  The difference was very subtle.  Roy’s comparison with a $7k Zenith PC reduces that device’s value.

6.       The ethernet and modified clock switch provides isolation, or filtering from upstream noise.  As Roy mentions, this has been known for some time.

7.       Switch versions with better clocks and power sources are undergoing rapid changes with the boutique audio companies.


@romaz why use the NUC as an endpoint when the sMS-200ultra or the ultraRendu provide the same functionality?  Was it because there wasn’t one of those devices to compare against the NUC?  Are the hard drives on those devices noisier than the eMMC on the NUC?  The sMS and uR can be powered by an LPS-1.2 or SR7/4 with less power.

I’d like to know how the TLS Linux OS compares to Roon ROCK when Roon is upsampling.  Likewise with HQPlayer.  Finally, both streaming to an endpoint.

1. On my Dual PC set up , on the renderer part i am running the Linux Roon OS on an SLC CF card with excellent SQ results significantly better than a SSD . I have a SOTM modified CF sata adapter  ( clock , regulators , capacity ) that i will receive in a few days and see if it brings better results .

 


@romaz " why use the NUC as an endpoint when the sMS-200ultra  "  - My experience between single PC and DUAL PC is very clear  , i always get better SQ with DUAL PC ( super micro board ) , i also found that the super micro board ( with sCLK clock board ) was superior to the SMS-200 ultra that i tried in my system .

By the way the sMS 200 ultra is using an SD card to host the os .

PCserver Supermicro X11SAA under Daphile  ,Jcat pcie net card ,Etherregen,e-red dock endpoint,powered by LPS 1.2 , SPS 500 , Sean Jacobs level 3 psu,  DAC Audiomat Maestro 3, Nagra Classic Amp , Hattor passive preamplifier , Martin Logan montis

Link to comment
9 hours ago, auricgoldfinger said:

 

The AA Cable Shield is made by the company that manufactures MuMetal.  The shield is made from Co-NETIC, which I believe uses the same alloys as MuMetal, but you may want to double check before buying.  The WBS-250 is $11.35/ft. and the WBS-375 is $12.60/ft.

Thank . The co-Netic is apparently  a proprietary alloy where mumetal is also an other proprietary alloy but more common source .

 

Where have you tried this Co-netic shield ? Which construction do you use  ( connected or not ? ,JSSG ? ....) ? Which SQ benefit did you get ?

 

Here in France there is a company HiFicable which is applying a double layer of mumetal very thin plasticized film to all his top of the line cable . The SQ benefit is usually a darker background .

PCserver Supermicro X11SAA under Daphile  ,Jcat pcie net card ,Etherregen,e-red dock endpoint,powered by LPS 1.2 , SPS 500 , Sean Jacobs level 3 psu,  DAC Audiomat Maestro 3, Nagra Classic Amp , Hattor passive preamplifier , Martin Logan montis

Link to comment
On 8/8/2018 at 8:29 AM, rickca said:

 

Yes, this uses a better board than my own as it uses a Celeron with a larger secondary cache which should result in less latency.  It also utilizes DDR4 RAM instead of DDR3.  At this point, I do not know which type of RAM is lower latency.

Link to comment
On 8/8/2018 at 8:44 AM, TheAttorney said:

Does the SOtM switch score 10 only if connected to the REF10? If so, what would the score be without the REF10?

I'm thinking that the TLS switch may still be a contender for those that don't want the extra cost and spaghetti count of an external reference clock.

 

The new SOtM switch is still the best switch among the switches that I have in my possession even without the REF10.  Without the REF10, I would rate it a 9.  The SOtM switch definitely isolates better than the other switches that I have.

Link to comment
On 8/8/2018 at 1:20 PM, Superdad said:

 

Hi Roy:

Welcome back!  I greatly enjoyed reading your novel-length report. Made me late getting into the office this morning! x-D

Of course of particular interest to us were your results that a good switch effectively made your upstream server tweaks mostly irrelevant.  Such bodes VERY well for our forthcoming from-the-ground-up EtherREGEN switch as among other things it is highly focused on producing a single, ultra-isolated Ethernet "output." 

 

Ours will also have exceptional clocking (on the clean side--with clock also passed back over the isolators to the "dirty" side), specially chosen Ethernet magnetics topologies (who knew there were so many choices!), and of course top-shelf LT3042/45 linear voltage regulation throughout.  We are using a SiLabs clock synthesizer--similar to the one that SOtM uses on all their clock boards--to generate the various clocks needed for the PHYs, the switch chip, etc.

 

While the reference clock for the synthesizer will be the venerable Crystek CCHD-575. (Production samples these days yield -110dBc/Hz @10Hz offset--lower than any production XO or OCXO under $450.) But our box (same case as UltraCap/UltraRendu) will also have an SMA jack to accept an external 10MHz reference clock such as the very fine Mutec REF10, so extremists like you can go "all the way." 9_9

 

[A bit off-topic but likely of interest to this audience:

One thing that many people do not realize is that the phase-noise/jitter on the clock outputs generated by the somewhat expensive SiLabs synthesizers can generally never be any lower than the XO or crystal used as driving reference.  The performance of the XO (or we suspect a custom ordered crystal) on the fine clock board that SOtM uses in so many of its products is not known and they have never published specs for it (as opposed to their impressively spec'd, $3,500 answer to the REF10).  I suspect, based on user reports, that the phase-noise of the board they use in their "Ultra" models falls somewhere a bit better than the Crystek 575 but far short of your REF10 or SOtM's new sCLK-OCX10.  Perhaps I am in-artfully saying this--I do so not to dig at a competitor.  I'm just trying to make clear the spectrum of performance--and the part in bold above was my main point in this aside.]

 

 

Thanks also for the rest of your report which I think puts a lot of things in perspective for readers.  Glad to hear that you still have your Mac minis to use as a touchstone.  It certainly is easy for folks to fall down the rabbit-hole of extreme optimizations with computers.  The differences can be very worthwhile, but it is important to get to the roots of the matter--so as to hopefully find simpler solutions, especially for those who would just like to plug'n'play.  Of course that is the goal of the small, optimized, Ethernet connected renderer devices (micro/ultraRendu, smS-200, Aries, Aurender, etc.).  Different paths for everyone of course!

 

Cheers,

--Alex C.

 

Hi Alex, I think we are all expecting nothing less than a home run with your switch.  It's good to know you will be including a master clock input.

Link to comment
On 8/8/2018 at 4:50 PM, soares said:

Any availability dates? Tks

 

If you're referring to SOtM's new switch, they are awaiting parts and so no official release date or price that I am aware of.  If you're referring to Uptone Audio's switch, I would like to know the answer to that one myself.

Link to comment
On 8/8/2018 at 6:03 PM, Adyc said:

Thanks Romaz for the excellent post. It seems to me like coming back to a full circle separating server and renderer into two machines. We have to give credit to Sonore and especially @barrows. They all from day one keep emphasizing putting server far away from renderer. That is also what I have been doing from day one. I will never consider putting a server near my hifi equipments. My power server is in my study room and far far away from my listening room. 

 

I think my point with separating server and renderer has more to do with distributing tasks (such as library management, DSP, upsampling, rendering, etc) so that a single CPU doesn't have to do all the work.  This should result in less latency, especially by the more important machine that is closest to the DAC.  If you have a fanless server and renderer, I wouldn't feel compelled to have to physically separate them into different rooms, at least not for SQ reasons.  The benefit of a switch, which Sonore is not advocating at all at this time, is that you now don't have to spend a lot of effort and money into optimizing your server.

Link to comment
On 8/8/2018 at 8:16 PM, lmitche said:

Here is a good article on the nuc model romaz referenced above. The review is written by an Intel employee. The loaded model with a 32gb emmc and 2gb of ram costs ~ $225.

 

https://techsterweb.wordpress.com/2017/01/18/arches-canyon/

 

Here is a way to get archlinux to run in ram:

 

https://www.ostechnix.com/ramroot-run-arch-linux-entirely-ram/

 

I haven't tried it, but have decent archlinux(audiolinux) experience and it looks very doable.

 

Tdp is 10 watts at 12 to 19 volts so a lt3045 power source is feasible for those of us without sr7 unobtanium. With ram and USB power a 2 to 5 amp solution would be best.

 

Stammheim here we come!

 

Thanks again Roy!

 

 

 

You're on the right track, Larry.  Without giving too much away, TLS's OS is their own variant of audiolinux.  The beauty of these small NUC boards is they are not difficult to power well.  I suspect 12V at 2A just might do it.

Link to comment

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now



×
×
  • Create New...