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

[rickca]

I'm still researching this, but it seems like AMD x470 provides at least one M.2 slot supporting NVMe PCIe V3 x4 straight to the CPU, so this may be an advantage of the AMD platform.

 

Also if you care about hard disk acceleration using a small SSD, AMD has this thing called StorMI which is AMD's answer to Intel RST caching.

https://www.amd.com/en/technologies/store-mi 

[romaz]

2 hours ago, numlog said:

Its also a choice between core speed or core/thread count and cache size, one of them has to be more important.

The latter appeals to my ''beliefs'' more, cores and cache being physical hardware differences, core speed being flexible with any CPU.

 

 

 

Forced to choose, I would agree with you.  Finding the perfect balance with just the right CPU is the goal and then from there, figuring out the best way to power it.

[romaz]

1 hour ago, rickca said:

I'm still researching this, but it seems like AMD x470 provides at least one M.2 slot supporting NVMe PCIe V3 x4 straight to the CPU, so this may be an advantage of the AMD platform.

 

Also if you care about hard disk acceleration using a small SSD, AMD has this thing called StorMI which is AMD's answer to Intel RST caching.

https://www.amd.com/en/technologies/store-mi 

 

Avoidance of a SSD is my chief goal since disk access isn't such a big deal with Linux.  Even a USB stick which is very slow sounds as good as an Optane drive with either Euphony or AL.  I think the performance of Optane is a bigger deal for Windows.

[R1200CL]

9 hours ago, drjimwillie said:

I think Alex suggested using a cable without a metal head, to prevent the possibility of magnetic coupling.  I use the Ghent with the metal Metz head.  What the best quality, non-metal Ethernet connector, please?  Maybe Ghent can use this head with the JSSG 360?

 

Ghent is ‘’perfect’’. Reason for Alex to say that was because most ethernet cables with metal plug follow the standard, and hence has the shield connected to the plug.

 

We designed the Gent ethernet cable not to have the shield connected to the plug. Reason to use metal plug was to hopefully add some shielding to the last mm. 

 

And Gent is using JSSG360. 

 

As an example Vodka use same principle, and not connecting the plugs to the shield. 

 

The only thing missing in that design is to use the Belden glued twisted pair cables. And add JSSG360. Which would made the cable very stiff. 

You could also use the special designed Belden plug.

 

Blue Jeans use a special Belden glued twisted pair cable made for them. 

 

 

[seeteeyou]

5 hours ago, numlog said:

what is appeal of the Intel 9900 CPU over upcoming Ryzen 3000 series? 

 

The flagship Ryzen 3000 will have 12 cores and 24 threads, bigger cache, probably at a lower cost too... to me it seems like the obvious choice.

 

IMHO 9900T should have an edge in terms of its 35W TDP and let's take a look at Ryzen 3000

 

https://wccftech.com/amd-ryzen-3000-cpu-specs-prices-online-retailer-ryzen-9-3850x-leak/

Quote

flagship 3850X = 135W TDP (16C32T)
3800X = 125W TDP (16C32T)
3700X = 105W TDP (12C24T)
3700 = 95W TDP (12C24T)
3600X = 95W TDP (8C16T)
3600 = 65W TDP (8C16T)

 

https://www.hdplex.com/hdplex-h3-v2-fanless-computer-chassis.html

Quote

H3 can use one side or both sides for CPU cooling depends on expansion card configuration and supports CPU with 80W Max TDP.

 

While we could still get away (depending on room temperature?) with an i7-8700K @ 95W TDP, it's hard to tell if both Ryzen 3700 and 3600X were good candidates for a fanless chassis. 3700X / 3800X / flagship 3850X should be out of the question while the safest bet is 3600.

 

OTOH, EPYC 3000 (i.e. embedded) seemed to be friggin' awesome since we're getting 32 PCIe lanes from both 3201 and 3251

 

https://en.wikichip.org/wiki/amd/epyc_embedded/3201

https://en.wikichip.org/wiki/amd/epyc_embedded/3251

https://www.supermicro.com/products/nfo/EPYC3000_Embedded.cfm

 

$705 for EPYC 3251

 

https://www.wiredzone.com/10028800

 

$540 for EPYC 3201

 

https://www.wiredzone.com/10028801

 

Block diagram here

 

https://www.supermicro.com/manuals/motherboard/EPYC3000/MNL-2172.pdf#page=18

 

Interesting read about the power-related stuff, time to talk to Ghent again?

 

https://www.supermicro.com/manuals/motherboard/EPYC3000/MNL-2172.pdf#page=21

Quote

The M11SDV-4C/4CT/8C/8CT/8C+-LN4F motherboard supports either 12V DC in power input or ATX power input. The 12V DC IN power input requires an 8-pin 12V DC IN connection to the JPV1 header. ATX power input requires an 8-pin 12V connections to JPV1 header and PJ1 a 4-pin power connector of an ATX power supply by power cable (24-pin ATX to 4-pin power cable for PJ1, Supermicro P/N: CBL-PWEX-1063). PJ1 allows motherboard control of the 5VStby, power on, power good, and ground signals from the ATX power supply.

 

ECC RAM might / might not make a difference, while we could actually try 4 Ethernet cables as mentioned before

 

https://audiophilestyle.com/forums/topic/30376-a-novel-way-to-massively-improve-the-sq-of-computer-audio-streaming/page/523/?tab=comments#comment-927951

https://audiophilestyle.com/forums/topic/30376-a-novel-way-to-massively-improve-the-sq-of-computer-audio-streaming/page/523/?tab=comments#comment-928056

[numlog]

So the real flagship is actually a 16 core, wow! 

There is also the NoFan 100W heatsink which might handle the 3800X depending on your requirements.

The NoFan 80W for 95W TDP 6600K  was also pushing it with heavier loads (DSD conversion),

 

However wont the 9900T be underclocked to achieve that low TDP?

If those extra cores really mattered another possibility could be to manually underclock a 3800X to within safe limits of the heatsink.

 

Btw how would you make use of all those PCIe lanes?

[rickca]

9 minutes ago, numlog said:

However wont the 9900T be underclocked to achieve that low TDP?

The base clock frequency is reported to be 1.7 GHz.

 

 

[Chopin75]

13 minutes ago, numlog said:

So the real flagship is actually a 16 core, wow! 

There is also the NoFan 100W heatsink which might handle the 3800X depending on your requirements.

The NoFan 80W for 95W TDP 6600K  was also pushing it with heavier loads (DSD conversion),

 

However wont the 9900T be underclocked to achieve that low TDP?

If those extra cores really mattered another possibility could be to manually underclock a 3800X to within safe limits of the heatsink.

 

Btw how would you make use of all those PCIe lanes?

Wouldn't  the 16 Core be overkill? I use 8-Core AMD Ryzen 2700x which only utilizes all cores if I do HQplayer PCM---> DSD256. Unless you are planning to do this and also M-ch surround and perhaps active downloading files at the same time within the same server that plays the music (single box), I don't think you need 16-core!!  

[numlog]

Ok, I gather the interest in this CPU is for it potential for external powering.

[shahed99]

15 hours ago, romaz said:

I'll venture a guess.  Your optimized server is better than your unoptimized NUC.

That seems to be the case. When You/Rajiv and others reported the finding in regards to NUC, it was mind blowing. Mostly because, how can an-off-the shelf hardwire can sound better than purpose built music server. The fact that NUC without any modification holds up so well to a purpose built music server is astounding.

 

15 hours ago, romaz said:

Put something like an SR7 or the TLS Reference 1 to the NUC and you may change your mind about the NUC sounding thin.

I wish I had spare PSU lying around to test these scenarios. It would have been very educational. I've no doubt the NUC would scale really well with great power supplies like that. 

 

15 hours ago, romaz said:

Also, if you're using an independently powered SOtM PCIe USB card in your server, this is obviously giving your server an edge over your unmodified NUC.  Try connecting your server to your DAC using one of standard integrated USB ports on your motherboard and the delta that you hear might give you some idea of how much better your NUC could sound once you send it to SOtM for clock replacement.

I've used my motherboards standard integrated USB port and the sound is frankly unlistenable. Goes to show, how good the USB output of the NUC is. However, not sure if the clock improvement is the only thing to close the delta. As the SoTM USB PCI-E has other things like filters and special voltage regulators. 

 

15 hours ago, romaz said:

As for WS2016 + AO sounding thicker, fuller and weightier than AL, the easiest way to find out in your system is just to boot your server into AL using a USB stick and compare.  They both can be tuned to sound fuller or leaner.  Regarding Euphony vs AL, yes, to my ears, Euphony sounds fuller and weightier and so it will cost you nothing to give Euphony a try but ultimately, if you're already happy with what you're getting from just your server, perhaps it's time to sell the NUC and stay with what you already have.  Finding the perfect balance is a tricky thing and what works for one person may not work for another but once you find that perfect balance, it's golden.

If I end up keeping my NUC, I will either use Windows Server on it (if possible) or try the Euphony. 

 

You and Rajiv reported that adding a NUC (with no clock modification) as a Roon endpoint to a high quality server like Innuos Zenith was an improvement in sound quality compare to using the Innuos alone. My custom built server I'm sure is not as good as Innuos. But still, it sounds better on its own comparing to AL/NUC endpoint. That has left my head scratching.

 

Many of you got great result with the unmodified NUC using it as Roon endpoint compare to stand alone server. That is not the case in my situation. That is why I'm in a dilemma doing the clock upgrade and better PSU no doubt will yield a better sound quality but whether the ROI would be worth it. I think I might try installing Windows Server on the NUC and see the result.

 

As a side question: May told me that it is possible to install sCLK-EX inside the NUC but would have to remove the 2 USB connection from the front of the case. Do you think there's added benefit of having a dedicated sCLK-EX inside rather than through USB Ultra?

[numlog]

19 minutes ago, Chopin75 said:

Wouldn't  the 16 Core be overkill? I use 8-Core AMD Ryzen 2700x which only utilizes all cores if I do HQplayer PCM---> DSD256. Unless you are planning to do this and also M-ch surround and perhaps active downloading files at the same time within the same server that plays the music (single box), I don't think you need 16-core!!  

Possibly, it might depend how you use them.

I still havent tested out if XXHighend is making use of the 1700's 8 Cores or tried out the ''core appointment schemes''.

Phasures PC comes in 12 core version, since XXHighend only allows PCM upsampling its never gonna be under the same kind of load as DSD conversion so I trust there is reason they recommend it beyond pure processing power.

there was also this thread

 

[motberg]

15 hours ago, Advieira said:

IMPRESSIONS WITH A NUC SERVER

 

Mikrotik router SFP Fiber + Nuc7i7BDNE Akasa (LPS Zerozone) > SnH-10G sclk-ex (stock ps) > Direcstream Sr. Bridge II

 

AudioLinux Ram: sound very sharp, bright, lean, very detailed, a little fatigating, very stable (nuc gets very hot)

Windows Server Data Center 2019 (no A.O) (M.2): details, bodied, harmonics richness, very enjoyable, network not very stable occasionally

 

I will keep AL for some auditions and impressing some friends, but for my ears, WS19 sounds better (understanding better as natural and real).

 

Now, I didn't understood why all that AL fuss...

 

Thanks for these comments

 

Anyone using or compare AL or Euphony with Volumio?

 

I had a chance to listen to a friends excellent DIY Player/DAC (using Volumio w/ X10SBA-L) in my stereo system (usually Win2012 core, AO, JPlay) and, even though the hardware was different - I got the impression that Volumio possibly offered software SQ on par with what I am used to, and in some respects (air and highs details) could be preferred by many folks... 

[motberg]

11 hours ago, austinpop said:

 

Yes, exactly. This was not a choice, as the H3 case does not have an ATX input in any case. So the only option is to snake the cable through the PCIe slot opening, and connect it directly to the motherboard.

 

Same here using a Thermaltake case, note the generous length of the HDPlex 400W ATX cable helps arrange this.

[motberg]

7 hours ago, romaz said:

 

Avoidance of a SSD is my chief goal since disk access isn't such a big deal with Linux.  Even a USB stick which is very slow sounds as good as an Optane drive with either Euphony or AL.  I think the performance of Optane is a bigger deal for Windows.

 

Anyone have suggestions on what type or characteristics a USB stick should have for persistent use with Euphony or AL ?

[romaz]

4 hours ago, Chopin75 said:

Wouldn't  the 16 Core be overkill? I use 8-Core AMD Ryzen 2700x which only utilizes all cores if I do HQplayer PCM---> DSD256. Unless you are planning to do this and also M-ch surround and perhaps active downloading files at the same time within the same server that plays the music (single box), I don't think you need 16-core!!  

 

You may be right but it's best not to assume.  More cores seem to matter even if you don't use HQP.  In my Roon server machine where I employ no upsampling or DSP, as I go from a 4-core i7 to a 6-core i7 (a 50% jump in cores), the improvement is notable in terms of a bigger and more dynamic sound.  No matter how high I crank up the 4-core i7's CPU frequency, I can't get it to match the 6-core i7's dynamics.  The 4-core sounds thin and small in comparison even with the 4-core i7 powered by a 19V SR7 and with the 6-core powered by an HDPlex.  At some point, adding more cores probably won't make a difference but until someone does the testing to find where that ceiling is, no one knows.  The advantage of parallel processing through multiple cores is that with software that is written to take advantage of those cores, tasks can be performed more efficiently and often more effectively and so ultimately, what is important is not how many cores but whether the software was written to take advantage of those cores.

 

A good example is Rob Watts' M-Scaler which utilizes a Xilinx XC7A200T FPGA and its 740 DSP cores.  This FPGA can upsample to 1 million TAPS while consuming only 12 watts.  There is no 12-watt X86 processor from either Intel or AMD that can achieve this level of performance.

[sunny_time_99]

2 hours ago, motberg said:

 

Anyone using or compare AL or Euphony with Volumio?

 

HI,

 

i am doing it for nearly 3 Weeks now. Still have no 100% favorite! that means in other words Volumio is damned good for being a Freeware. it depends on the Hardware, which System is the best. On my older futro 900-fanless PC daphile wins every contest. ... so there is no operating System that wins every contest on all hardware-Setups. After listening to Volumio on a single NUC7i7DNHE (i own two in Plato Housing for better A/B-Test) for nearly two weeks i changed back to euphony yesterday - and back to volumio, becouse i missed the higher Dynamics from Volumio.

 

•  

•  
•  
•  #13764

Posted 1 hour ago

9 hours ago, romaz said:

 

Avoidance of a SSD is my chief goal since disk access isn't such a big deal with Linux.  Even a USB stick which is very slow sounds as good as an Optane drive with either Euphony or AL.  I think the performance of Optane is a bigger deal for Windows.

 

Anyone have suggestions on what type or characteristics a USB stick should have for persistent use with Euphony or AL ?

I was comparing optane, USB-Stick 2.0, SLC USB-Stick this Weekend with Euphony. On Optane, Euphony sounds a bit thin and bloodless to my ears. A USB 2.0 Intenso Stick was on the same Level - so no benefit. A seperate powered  SLC-USB-Stick from Winkom opened the door to never heared subsonics.

Regards

Sunny

[romaz]

5 hours ago, shahed99 said:

That seems to be the case. When You/Rajiv and others reported the finding in regards to NUC, it was mind blowing. Mostly because, how can an-off-the shelf hardwire can sound better than purpose built music server. The fact that NUC without any modification holds up so well to a purpose built music server is astounding.

 

My initial observations with the NUC were surprising for a couple of reasons.  First, that a $120 Celeron-based NUC + AL could outperform (to my ears) products like a microRendu/ultraRendu/sMS-200ultra that were designed from the ground up for audio playback with high level clocking and ultra low noise voltage regulators.  Second, that this same $120 NUC + AL when powered by a DR SR7 could outperform (to my ears) a $7k Innuos Zenith SE.  

 

This was my first inkling that low power CPUs like an ARM processor are probably not optimum for an endpoint.  For a cheap unoptimized Celeron to outperform these highly regarded endpoints, I could offer no other explanation and this theory was only bolstered when I moved to an i7 NUC and found that the gap had widened further.  It also suggested the OS matters and it matters a lot.  While both my initial NUC and the Zenith SE utilize Celerons and with both powered by high quality PSUs, I believed that it was the OS that was the difference maker and I recall that Rajiv found this to be his observation also, that his SE when booted to AL resulted in improvement.

 

Of course, optimizing the NUC by replacing its clocks has resulted in further significant improvements but the lesson that the unoptimized NUC taught me is that the performance that a CPU brings, the quality of the OS, and of course, the quality of the PSU matter just as much if not more.  With respect to your server, these qualities have already been considered and so I'm not surprised that your unoptimized NUC powered by an sPS-500 isn't meeting your expectations.

 

As for ROI, replacing clocks on your NUC is the last thing I would do.  Figure out a way to better power your NUC.  This would be my first priority.  Second, because it's easy enough to do, go ahead and give Euphony a try just in case it's signature is more to your liking.  The trial is free and it is easy to do.  Or else, give WS 2016/AO a try on the NUC but of course, you won't be able to boot it from a USB stick and so you'll have to install either an SSD or preferably an Optane card.  Ultimately, if you're already happy with your optimized server, your best ROI might be to sell your NUC and cut your losses.  

[romaz]

2 hours ago, motberg said:

 

Anyone have suggestions on what type or characteristics a USB stick should have for persistent use with Euphony or AL ?

 

My experience is that the quality of the USB stick has no impact on SQ.  I've tried an expensive SLC USB stick I had lying around and it resulted in no change in SQ that I could hear with Euphony compared against a cheap $5 Sandisk.  Of course, if you are using AL ramrooted, for sure the USB stick won't matter.

 

Just get a fast USB 3.0 or 3.1 USB stick to keep your wait times to a minimum during bootup.  These devices have limited write cycles before they fail and so make sure you have a backup.  Fortunately, these things are cheap.

 

This will be my last post on AS for the foreseeable future.  Please forgive me if I don't respond to queries that are addressed to me.  Best wishes and I hope to meet up with some of you in Munich in May.

[sunny_time_99]

40 minutes ago, romaz said:

 

My experience is that the quality of the USB stick has no impact on SQ.  I've tried an expensive SLC USB stick I had lying around and it resulted in no change in SQ that I could hear with Euphony compared against a cheap $5 Sandisk.  Of course, if you are using AL ramrooted, for sure the USB stick won't matter.

 

Just get a fast USB 3.0 or 3.1 USB stick to keep your wait times to a minimum during bootup.  These devices have limited write cycles before they fail and so make sure you have a backup.  Fortunately, these things are cheap.

 

This will be my last post on AS for the foreseeable future.  Please forgive me if I don't respond to queries that are addressed to me.  Best wishes and I hope to meet up with some of you in Munich in May.

 

thats right, The Magic comes when you power the usb-stick with a seperate and high quality power supply.

 

use  e.g. a usb 3.0 Adapter, take out the ground and V+ line, cut V+ line, connect V+ and gnd from your psu... then you will be able to hear really big differences between CF, SD, SLC and simple blue jeans usb sticks. These differences are much bigger than clock tunings. Everything that ist connected to the nuc/pc must have a seperate power supply and the V+-Line cut. thats a basic in a audiophile Computer audio setup.

 

regards

 

sunny

[FelipeRolim]

22 hours ago, Advieira said:

IMPRESSIONS WITH A NUC SERVER

 

Mikrotik router SFP Fiber + Nuc7i7BDNE Akasa (LPS Zerozone) > SnH-10G sclk-ex (stock ps) > Direcstream Sr. Bridge II

 

AudioLinux Ram: sound very sharp, bright, lean, very detailed, a little fatigating, very stable (nuc gets very hot)

Windows Server Data Center 2019 (no A.O) (M.2): details, bodied, harmonics richness, very enjoyable, network not very stable occasionally

 

I will keep AL for some auditions and impressing some friends, but for my ears, WS19 sounds better (understanding better as natural and real).

 

Now, I didn't understood why all that AL fuss...

 

Exactly what I said here:

 

[Montage]

As others have reported, I found AL to be bright sounding on my NUC7i7DNHE in Akasa case with UpTone JS-2 linear power supply (12V) acting as a Roon endpoint. Has anyone directly compared the JS-2 at 12V with a HDPLEX a 19V on a NUC7i7DNHE?

 

On the NUC endpoint I switched to Windows Server 2019 with Audiophile Optimiser 3 (Beta) and Roon Bridge and have been very happy with the resulting sound via ISO Regen to Pacific dac. Following the advice here from Romaz and others I've just ordered an Optane M2 for the operating system (currently on SSD), so I'm hoping for a further step-up.

 

I prefer AL on my Roon server compared to Roon ROCK. I've not yet tried Server 2019 and AO3B on the server, it will have to wait for an Optane compatible mobo. However, I will give Euphony a try.

 

Thanks to everyone here for the on-going advice!

[Johnseye]

6 hours ago, romaz said:

 

My initial observations with the NUC were surprising for a couple of reasons.  First, that a $120 Celeron-based NUC + AL could outperform (to my ears) products like a microRendu/ultraRendu/sMS-200ultra that were designed from the ground up for audio playback with high level clocking and ultra low noise voltage regulators.  Second, that this same $120 NUC + AL when powered by a DR SR7 could outperform (to my ears) a $7k Innuos Zenith SE.  

 

This was my first inkling that low power CPUs like an ARM processor are probably not optimum for an endpoint.  For a cheap unoptimized Celeron to outperform these highly regarded endpoints, I could offer no other explanation and this theory was only bolstered when I moved to an i7 NUC and found that the gap had widened further.  It also suggested the OS matters and it matters a lot.  While both my initial NUC and the Zenith SE utilize Celerons and with both powered by high quality PSUs, I believed that it was the OS that was the difference maker and I recall that Rajiv found this to be his observation also, that his SE when booted to AL resulted in improvement.

 

Of course, optimizing the NUC by replacing its clocks has resulted in further significant improvements but the lesson that the unoptimized NUC taught me is that the performance that a CPU brings, the quality of the OS, and of course, the quality of the PSU matter just as much if not more.  With respect to your server, these qualities have already been considered and so I'm not surprised that your unoptimized NUC powered by an sPS-500 isn't meeting your expectations.

 

As for ROI, replacing clocks on your NUC is the last thing I would do.  Figure out a way to better power your NUC.  This would be my first priority.  Second, because it's easy enough to do, go ahead and give Euphony a try just in case it's signature is more to your liking.  The trial is free and it is easy to do.  Or else, give WS 2016/AO a try on the NUC but of course, you won't be able to boot it from a USB stick and so you'll have to install either an SSD or preferably an Optane card.  Ultimately, if you're already happy with your optimized server, your best ROI might be to sell your NUC and cut your losses.  

 

I'm going back a couple years, but am curious about your discovery path and reasoning.  Roon's recommendation has always been more cores and larger cache.  You started looking at i7's but then moved towards the Celeron a couple years ago. The focus on keeping the lowest possible TDP hasn't changed, but you were focusing on lowering the frequency.  What was your thinking back then and did you compare an i7 against the Celeron?

 

[Johnseye]

Are those who have compared and prefer Euphony over AL, using the Stylus player, Roon alone, HQPlayer alone or Roon + HQPlayer?  I'm wondering if the perceived improvement is due to the song being loaded into RAM with the Stylus player or something else.

[austinpop]

56 minutes ago, Johnseye said:

Are those who have compared and prefer Euphony over AL, using the Stylus player, Roon alone, HQPlayer alone or Roon + HQPlayer?  I'm wondering if the perceived improvement is due to the song being loaded into RAM with the Stylus player or something else.

 

I'm in the process of evaluating Euphony. I'm trying it on both the server and the endpoint. 

 

Due to the Roon Bridge bug that @bobfa mentioned, I'm running Roon Core on the server, and Squeezelite on the endpoint. SQ is definitely different, and it has nothing to do with Stylus. I'll say more after further listening.

[Chopin75]

55 minutes ago, austinpop said:

 

I'm in the process of evaluating Euphony. I'm trying it on both the server and the endpoint. 

 

Due to the Roon Bridge bug that @bobfa mentioned, I'm running Roon Core on the server, and Squeezelite on the endpoint. SQ is definitely different, and it has nothing to do with Stylus. I'll say more after further listening.

If u are running roon then u are not using stylus correct?  While running for comparision you should also specify what music files u use - 24/96, or 24/352 or dsd256 etc. i find stylus does best with super hi res.  For stylus, u should also specify where the music files are - in the local ssd where euphony OS is or external drive. For exernal drive, ecache or not?  100% buffer to get entire music file in RAM.