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

[seatrope]

23 hours ago, romaz said:

 

With AudioLinux, I was forced to cap an 8700K at 3.8GHz because any speeds beyond this sounded harsh and so this supports Jord's comment.  But with Euphony and Stylus, I let the 8700K run with no cap whatsoever and notice that CPU frequency will typically reach 4.4-4.5GHz consistently, however, I get no harshness at all.  In fact, the higher the frequency a CPU is capable of, the better, and so I imagine the ultimate CPU at this time would be something like an i9-9900K that provides 8-cores and a max frequency of 5GHz and a TDP of only 95w.  I hope to be able to test this soon but what I will say is that with Euphony + Stylus on a single box 8700K machine, my reclocked i7 NUC driven by a 19V SR7 sounds absolutely puny.  Even with the 8700K server powered by an HDPlex 400W ATX LPSU and with no special clocking outside of my tX-USBultra, I prefer the big server to the i7 NUC but as I have figured out a way to independently power both the 8700K CPU and ATX motherboard (using a DC-ATX converter) with SR7 rails, the i7 NUC has now been officially retired.

 

@romaz Thanks for your insights here.

 

FYI, if you have not read, there's been quite the debacle about the i9-9900K and it's "TDP" of 95W. There are numerous articles on the web about how this is quite disingenuous on Intel's part. The real TDP is more like 140W which precludes us running this processor in any of the HDPLEX chassis. If you really limit the 9900K to 95W the speed is more in the range of 4.1- 4.3GHz (excuse my poor recollection) and is about the same speed as a much cheaper 2700X Ryzen. 

 

Yeang

[austinpop]

23 minutes ago, Bruce Orr said:

@austinpop  So Rajiv, what is the server you have embedded Euphony + Stylus into?  Is it the one you had built for yourself?

 

Hi Bruce,

 

I described it in the index. The first link is my server. The second is Roy's magnum opus, where he describes his server journey. 

• @austinpop custom server update
• @romaz System update March 2019

I did not build this server, although I like to think I could do it. My server is the first one Roy built for himself, and then sold to me after he decided to go with the gaming Z390 mobo and 8700K CPU.

[tapatrick]

5 hours ago, bobfa said:

Seems like we need a "current summary of findings". 

Yes please!

 

[romaz]

23 hours ago, Nenon said:

@romaz Good to hear back from you. I also found a single Euphony installation running Stylus to sound best in my system. 

I've read most of your posts and know that you don't like to use SSD drives, even powered by a dedicated DC rail and good quality SATA cable, which I remember you had good feedback for. The computer I use does not support Optane drives. I could only pick between SSD or HDD for the Euphony OS. Unless I can keep the HDD outside of the chassis it's not an option for me. I would not want those vibrations inside my chassis. So the most logical way is to use SSD. And this is what I have done.

I am assuming the noise generated from the SSD drive is at its strongest levels when there is some type of I/O activity. Stylus caches the track in memory before playing. That means the SSD drive is not used for music retrieval while playing. The Linux OS could potentially use it, but a stripped down Linux version like the one Euphony uses would not require a lot of I/O activity.

So my question really is, how bad is to use a SSD drive for a single box Euphony installation? Is the negative impact of the SSD diminishing as Stylus is loading the track in memory before playing? 

While I have other options for the my music drive (USB, NAS, etc.), I still need to load the OS from a hard drive. I don't find using a USB stick ideal - prefer to keep the USB for the solely use of my DAC.

 

 

The concept of electrical noise generated by an SSD was first brought to public consciousness, I believe by Paul Pang, and he was also the first that I'm aware of to suggest using alternative lower noise media like USB flash drives and slower SATA II SLC SSDs to run the OS.  Many, including myself, have found Paul's findings to be true in our own systems.  Marcin's upcoming JCAT server looks like it will be based on SLC media.  Keetakawee Punpeng, creator of Fidelizer and the Nimitra server will offer SLC media in his best custom builds also.  

 

While noise generated by the OS drive is an important consideration, so is performance, especially with Windows which engages in high amounts of disk I/O.  This is where NVMe drives and their very low latencies have an advantage over slower drives including any drives connected to the motherboard via SATA.  With Windows, these very low latency drives result in better immediacy although this is much less relevant with either AL or Euphony.  The advantage of an Optane drive is that it runs on this very low latency NVMe bus while also being a very low noise drive and so it encompasses the best of both worlds.

 

I have recently been educated that modern motherboards that have M.2 NVMe slots that aren't necessarily "Optane ready" (i.e. AMD boards) can still utilize an Optane card as a storage (or OS) drive but not as a cache drive, which is what Optane was originally designed for.  This means that if your motherboard has an M.2 drive capable of NVMe, you are probably capable of using an Optane drive.  It should be easy enough to try it.

 

If Optane is not possible, a USB stick is still not a horrible way to go with Euphony and or Stylus.  While not optimal, according to Euphony, there is no SQ hit with using a USB stick for the OS and I think this is because Euphony and many Linux-based OS's have small enough footprints that they reside mostly in RAM by default and so disk I/O is nowhere close to Windows. Sonically,  I have confirmed this to be the case in my own server meaning that I don't detect a meaningful SQ difference between an Optane drive and a USB stick as an OS drive.  Of course, if you have a large library and you are interested in keeping your files within your server's internal drive, then you have no choice but to use large capacity storage media like an SSD or hard drive.  At that point, see what you get and if you aren't bothered by it, that's all that matters.

[romaz]

22 hours ago, Dev said:

I just need to decide on the motherboard and power supply (which seems the most critical and complex part).  

 

Agreed.

[romaz]

21 hours ago, lateboomer said:

The essence of this is how to independently power the cpu? Could someone provide details guide? Could I also independently power cpu at intel NUC ? Thank you.

 

The original ATX spec called for a 12V rail for the CPU and 12V, -12V, 5V, -5V, 5VSB, and 3.3V for the motherboard and so lots of rails although most modern ATX motherboards no longer use the -12V and -5V rails (now considered legacy rails) and if you never go into standby mode, you can get by without a 5VSB rail. 

 

So, with any modern motherboard that accepts ATX power, you will generally see a multipin ATX connector (usually 20 or 24-pin) and a multipin CPU connector (usually 4-8 pins).  It is this flexibility to power the CPU separately that is the difference maker and what disqualifies NUCs and any motherboard with an embedded CPU for any server I might build moving forward since these types of motherboards using take a single DC input.  This was discussed at some length on the other thread back in 2017 and it is one of the big reasons the Innuos Zenith SE sounded better than my own 2017 server build. 

 

What is interesting is when I first tried the HDPlex DC-ATX converter back in 2017, I was not impressed by what I heard with this converter powered by a 19V rail from my SR7 although back then, I used this converter to power both the motherboard and a Xeon CPU.  I think it was just too much to ask.  This time around, I used this converter to power just the motherboard and with a special cable made for me by Ghent, I'm using a 12V rail from my SR7 to independently power the CPU and this has made all the difference.

 

This is the HDPlex DC-ATX converter I am presently using:

 

https://www.hdplex.com/hdplex-400w-hi-fi-dc-atx-power-supply-16v-24v-wide-range-voltage-input.html

[romaz]

12 hours ago, TheAttorney said:

Every time I'm ready to finally upgrade my laptop to a supa dupa new server, the conversation moves on again to something even more wonderful! Do you guys never stop? 🙂

 

With the recent talk of m.2 Optane drives and a single box Euphony server, I'm wondering if I can truely keep it simple by putting all my music files and OS on a single optane drive in a single NUC?

 

My compressed redbook FLAC files currently total under 180GB, and are unlikely to grow much more in the new streaming world.

And one can readily buy optane drives up to 500GB+ (although I don't know if the larger sizes will physically or logically work in a NUC environment). So what would stop this being a good idea?

 

Your situation is becoming more the norm than the exception which is why Optane drives seem more viable today.  Some motherboards have multiple M.2 slots as well.

[Nenon]

On 5/23/2019 at 2:50 PM, Nenon said:

When you buy Euphony, it only has one license. That's the license you need to get.

There are different ways how you can configure Euphony after you get a license. You would have to try for yourself what works best, but so far Roon + StylusEP seems to be the winner for Roon users.

 But you can get a 30 day trial on a USB stick before you buy it.

 

FYI - this is no longer true. There is a EuphonyEP license now as reported earlier. 

[romaz]

1 hour ago, seatrope said:

@romaz Thanks for your insights here.

 

FYI, if you have not read, there's been quite the debacle about the i9-9900K and it's "TDP" of 95W. There are numerous articles on the web about how this is quite disingenuous on Intel's part. The real TDP is more like 140W which precludes us running this processor in any of the HDPLEX chassis. If you really limit the 9900K to 95W the speed is more in the range of 4.1- 4.3GHz (excuse my poor recollection) and is about the same speed as a much cheaper 2700X Ryzen. 

 

Yeang

 

I fully get what you're saying.  Both the 8700K and 9900K are based on a 14nm die and yet a 9900K with 2 more cores and 300MHz of greater headroom is supposed to generate the same heat?

 

However, issues you are reading about don't have to apply for audio.  I realize that with the Pink Faun 2.16, based on how this machine can be configured as an upsampling machine using HQP and AL and based on what others have posted, I have read claims that this machine consumes 100 watts at a minimum and functions as a space heater.  

 

In my case, I do no upsampling since I have a separate Chord M Scaler that does this for me.  With Euphony + Stylus and with the 8700K CPU unrestrained, as I look at my Euphony meter, CPU load across 6-cores and 12-threads is consistently <1-2% during music playback and CPU frequency peaks as high as 4.63GHz for a given core.  The CPU cores run between 38-47 degrees C even though I have been playing music continuously for the past 3 hours.  During boot up, my Kill-A-Watt meter climbs as high as 92 watts of consumption for about a split second but once in a steady state, while playing music, this machine consumes no more than about 50 watts and as little as 38 watts.  The HDPlex chassis handles all of this well (it barely feels warm to the touch even with the machine kept on 24/7 for most of the past week) in my enclosed but ventilated cabinet.  I would be shocked if my setup can't handle the 9900K.

 

The AMD 1800X also has a rated TDP of 95 watts and while it has 8-cores, it has a max frequency of only 4GHz and so this is not the best case scenario for Stylus.  If rumors prove true, the upcoming AMD 3000 series CPUs with a die of only 7nm may be the ultimate CPU.  The rumored 3800X is purported to have 12-cores and a max CPU frequency of 5GHz while sporting a TDP of only 105 watts.  This TDP may be a big deal for a Pink Faun machine running AL and HQP but I suspect it won't be a big deal for Euphony + Stylus.

 

As for motherboards, this matters and I'm hoping the new crop of mini-ITX AMD X570-based motherboards will be better than what's available.  Because of the unnecessarily large fan mounting posts that AMD CPUs use, they use up precious space on the small mini-ITX motherboards making it challenging for the motherboard manufacturers to incorporate high level VRMs.  In this sense, if one is looking for a mini-ITX board, Intel boards are far superior than any of the AMD boards that are presently available and the best board I have so far found is the Asrock Z390 Phantom Gaming ITX-ac.  For a solid AMD board, it seems you really have to go full ATX.

[austinpop]

Mindful of @bobfa's point that these general topics are OT in the Shootout thread, I'm posting this here.

 

1 hour ago, romaz said:

This TDP may be a big deal for a Pink Faun machine running AL and HQP but I suspect it won't be a big deal for Euphony + Stylus.

 

If there is one "new" finding in all of this recent progress, it is that combined with all the previously known optimizations of clean power, separate CPU power, linear supplies, better clocks, etc., running with a very powerful CPU like the i7-8700K provides a level of dynamism and clarity we have not heard from the low-powered CPU approach of the likes of the ZENith SE, the sMS-200ultra, and even the NUCs.

 

However the key thing here is that we operate these massively powerful CPUs at very low utilization - as we see with the non-upsampling use cases of Roon or Stylus.

 

Once you add upsampling into the mix, all bets are off, as it remains to be seen what the effect of that energy consumption due to high CPU utilization will be on:

1. the ability of fanless cases to cope with the added heat
2. the effect CPUs running at very high temps on SQ.

I had a chat with Nuno @nvitorino about this at Munich. I asked him if he would consider incorporating our findings in his server line - i.e. something like a ZENith or Statement containing powerful i7 and i9 cores. He said it was OK for us hobbyists to go this route, as we have the skill and knowledge to control our workloads and ensure we ran at very low utilizations. As a commercial vendor, he cannot control what people do with the box, and his concern is that people who try to do heavy DSP and upsampling may not get the SQ benefit - indeed, they may get the opposite effect. Also, this would put a strain on the PSUs, since they would have to design the PSUs for the heaviest case, which would require far more powerful and expensive PSUs than are currently in those boxes.

 

So this is a valid concern.

 

I think folks in the upsampling HQPlayer space need to run their own experiments and see where the sweet spot is. I would not be surprised if the answer is not standalone, but a distributed configuration: a separate powerful server for upsampling, and an endpoint, perhaps still using a powerful i7 or i9 core, but running NAA at very low CPU utilization. Please note - this is pure conjecture on my part, but it's an educated guess on what I and Roy have experienced thus far with non-upsampled Roon and Stylus.

[sandyk]

23 minutes ago, austinpop said:

Mindful of @bobfa's point that these general topics are OT in the Shootout thread, I'm posting this here.

 

 

If there is one "new" finding in all of this recent progress, it is that combined with all the previously known optimizations of clean power, separate CPU power, linear supplies, better clocks, etc., running with a very powerful CPU like the i7-8700K provides a level of dynamism and clarity we have not heard from the low-powered CPU approach of the likes of the ZENith SE, the sMS-200ultra, and even the NUCs.

 

However the key thing here is that we operate these massively powerful CPUs at very low utilization - as we see with the non-upsampling use cases of Roon or Stylus.

 

Once you add upsampling into the mix, all bets are off, as it remains to be seen what the effect of that energy consumption due to high CPU utilization will be on:

1. the ability of fanless cases to cope with the added heat
2. the effect CPUs running at very high temps on SQ.

I had a chat with Nuno @nvitorino about this at Munich. I asked him if he would consider incorporating our findings in his server line - i.e. something like a ZENith or Statement containing powerful i7 and i9 cores. He said it was OK for us hobbyists to go this route, as we have the skill and knowledge to control our workloads and ensure we ran at very low utilizations. As a commercial vendor, he cannot control what people do with the box, and his concern is that people who try to do heavy DSP and upsampling may not get the SQ benefit - indeed, they may get the opposite effect. Also, this would put a strain on the PSUs, since they would have to design the PSUs for the heaviest case, which would require far more powerful and expensive PSUs than are currently in those boxes.

 

So this is a valid concern.

 

 

Hi Rajiv

 I agree 100% with the importance of all these things, and note that several years ago in Hi Fi Critic magazine they also reported that the faster processors resulted in improved performance.

 Perhaps their noise profiles are pushed further up the Spectrum ?

As you say, heat becomes the main problem with smaller and affordable cases.

Kind Regards

Alex

 

 

[Johnseye]

2 hours ago, austinpop said:

Mindful of @bobfa's point that these general topics are OT in the Shootout thread, I'm posting this here.

 

 

If there is one "new" finding in all of this recent progress, it is that combined with all the previously known optimizations of clean power, separate CPU power, linear supplies, better clocks, etc., running with a very powerful CPU like the i7-8700K provides a level of dynamism and clarity we have not heard from the low-powered CPU approach of the likes of the ZENith SE, the sMS-200ultra, and even the NUCs.

 

However the key thing here is that we operate these massively powerful CPUs at very low utilization - as we see with the non-upsampling use cases of Roon or Stylus.

 

Once you add upsampling into the mix, all bets are off, as it remains to be seen what the effect of that energy consumption due to high CPU utilization will be on:

1. the ability of fanless cases to cope with the added heat
2. the effect CPUs running at very high temps on SQ.

I had a chat with Nuno @nvitorino about this at Munich. I asked him if he would consider incorporating our findings in his server line - i.e. something like a ZENith or Statement containing powerful i7 and i9 cores. He said it was OK for us hobbyists to go this route, as we have the skill and knowledge to control our workloads and ensure we ran at very low utilizations. As a commercial vendor, he cannot control what people do with the box, and his concern is that people who try to do heavy DSP and upsampling may not get the SQ benefit - indeed, they may get the opposite effect. Also, this would put a strain on the PSUs, since they would have to design the PSUs for the heaviest case, which would require far more powerful and expensive PSUs than are currently in those boxes.

 

So this is a valid concern.

 

I think folks in the upsampling HQPlayer space need to run their own experiments and see where the sweet spot is. I would not be surprised if the answer is not standalone, but a distributed configuration: a separate powerful server for upsampling, and an endpoint, perhaps still using a powerful i7 or i9 core, but running NAA at very low CPU utilization. Please note - this is pure conjecture on my part, but it's an educated guess on what I and Roy have experienced thus far with non-upsampled Roon and Stylus.

 

Your assumptions are correct Rajiv.  I do not own a Chord M Scaler. My direction has been to use an NOS DAC and HQPlayer. As you know, last year I was using a single ASRock ITX board modified by SOtM and powered by the SR7 with the HDPlex converter. Separately powering the board and CPU by separate rails of the HDPlex 400 LPSU was actually an improvement. 

 

Most recently I have incorporated a SOtM modified NUC as the endpoint. It's using the sCLK-EX taps of an sNH-10G and both are powered by the SR7. Everything is clocked by the Mutec; server, NUC and switch. The NUC is running as an HQPlayer NAA.

 

I upsample with HQPlayer and depending on the filter can exceed the tap count of an M Scaler. I don't question the M Scaler's prowess but there's pros and cons to both solutions. Since running my system this way I have been challenged to differentiate analog from digital. I will take analog sourced vinyl and put it side by side with a digital source and I can't tell the difference. This is incredibly easy to blind test. I had Bob over last week and he confirmed what I had.  He couldn't tell the difference either. 

 

On another note, I have finished correcting my room after Jeff Hedback's testing, analysis and recommendations. The change is remarkable. This kind of effort is limited to those who can make changes to their listening environment. Depending where this is, the wife factor can be in play but there are still tasteful and wife acceptable changes that can be made. In my case I had free reign and so applied absorbers and diffusers on the ceiling and walls. The impact is bigger than all the little tweaks we make. It's at the macro level. 

 

[lateboomer]

8 hours ago, Dev said:

 

This is what I have been looking for as well - one slot for USB and another for Ethernet. Though I am using Wifi at the moment, but I would not like to loose the flexibility going forward. Mobo with 2 PCIe slot would require to step up to the m-ATX motherboard. I am also waiting for the release of i9-9900 in couple of months.

I wonder for the Asrock H370M-ITX/ac mobo, is it doable to find a splitter to split up the pcie slot to 2 pcie slots to use with jcat femto net card and also jcat USB femto card? Or is it such thing to convert the M.2 key E wifi connector to additional pcie slot to use with USB card? 

[seeteeyou]

There's yet another alternative to HDD/SSD/USB sticks, IIRC Roy was quite happy about eMMC when he's testing his very first NUC back then

 

64GB for $35 + $5

https://store.pine64.org/?product=64gb-emmc

https://store.pine64.org/?product=usb-adapter-for-emmc-module

 

64GB for $22 + adapter

https://www.futureelectronics.com/p/7113132

https://www.kingston.com/en/embedded/emmc

https://www.eciaauthorized.com/en/search/kingston/EMMC

 

128GB for $59 + $10

https://forum.odroid.com/viewtopic.php?f=53&t=2725

https://www.hardkernel.com/shop/usb3-0-emmc-module-writer/

https://www.ebay.com/itm/eMMC-Module-C2-Android-Green-Box/223284605230

https://www.ebay.com/itm/eMMC-5-1-Module-XU3-XU4-Android-White-Dot/223284614850

 

 

 

 

 

 

 

BTW, 2nd gen of Xeon Silver (4215 with 85W TDP) looked fairly interesting for those of us who are willing to pay the premium and go for an ATX fanless case

 

https://en.wikichip.org/wiki/intel/xeon_silver/4215

https://ark.intel.com/content/www/us/en/ark/products/193389/intel-xeon-silver-4215-processor-11m-cache-2-50-ghz.html

 

About $800 at the moment

 

https://www.provantage.com/hpe-p02598-b21~7CMPTCAJ.htm

https://www.provantage.com/hpe-p02494-b21~7CMPTCCW.htm

https://www.provantage.com/intel-cd8069504212701~7ITEP6HP.htm

 

An ATX motherboard for $330

 

https://www.provantage.com/supermicro-mbd-x11spl-f-b~7SUPM5QQ.htm

https://www.wiredzone.com/supermicro-components-motherboards-single-processor-x11spl-f-41110336

 

128GB Optane DC for $680 versus 128GB DDR4 for $1390

 

https://www.avadirect.com/128GB-Optane-DC-Persistent-Memory-Module/Product/12697853

https://www.amazon.com/Supermicro-MEM-DR412L-HL01-LR26-1x128GB-Reduced-LRDIMM/dp/B07FTWJBKJ

 

Obviously the frequency could only reach 3.5GHz in order to keep the TDP below 90W, it would be interesting to compare that with an i9-9900K IMHO.

 

OTOH, €20,000 SGM Extreme Server actually incorporated dual Xeon processors and it would be fun to give that a try with a relatively affordable ATX motherboard

 

http://www.acmemicro.com/Product/15819/Supermicro-X11DPL-i-Motherboard-Intel-Xeon-Scalable-Dual-Socket-P-

https://www.newegg.com/supermicro-mbd-x11dpl-i-o-2nd-generation-intel-xeon-scalable-processors-intel-xeon-scalable-proces/p/1B4-005W-00149

[seeteeyou]

While passive Thunderbolt 3 cables must be short enough (e.g. half a meter) in order to keep up with the 40Gbps bi-directional data transmission, we could consider adding a dock to ASRock Z390 Phantom Gaming-ITX/ac

 

https://www.amazon.com/Sonnet-Express-Thunderbolt-Expansion-ECHO-EXP-SE1-T3/dp/B01MTP2XMZ

https://www.amazon.com/Sonnet-Echo-Express-III-D-Thunderbolt/dp/B0713SX7VV

 

 

 

 

[lateboomer]

Thank you. Very impressive product. 

[seeteeyou]

I've gotta clarify that H370M-ITX/ac didn't come with Thunderbolt 3 in the first place and therefore it won't work with those products from Sonnet. In order words, it's necessary to replace the motherboard with Z390 Phantom Gaming-ITX/ac with built-in Thunderbolt 3.

 

However, we have to be aware that most likely all 16 processor-connected PCIe lanes are occupied by the single PCIe x16 slot and therefore Thunderbolt 3 (as well as M.2 slot?) should be going through PCH.

[lateboomer]

Another thing is for single PC setup running Euphony Stylus, if os should be run in Optane card then where should music files be stored in optimized way?  Is it ssd or hdd? What is the type. I tried to skim through the index but cannot find the info. 

[seeteeyou]

Assuming that RAM drives should be the very best for music storage in terms of latency, Optane (occupying processor-connected PCIe lanes) seemed to come pretty close according to Rajiv

 

https://audiophilestyle.com/forums/topic/30376-a-novel-way-to-massively-improve-the-sq-of-computer-audio-streaming/page/462/?tab=comments#comment-900522

 

Some folks actually managed to edit the BIOS and PCIe bifurcation was working afterwards, then Optane could be connected to the processor of an ITX motherboard directly

 

https://hardforum.com/threads/pcie-bifurcation.1870298/page-16

 

A few links to some risers, in theory we could have up to 4 processor-connected PCIe slots if there were a (fanless) case that's gonna fit

 

https://peine-braun.net/shop/

http://www.ameri-rack.com/ARC2-PELY423-C7_m.html

https://hardforum.com/threads/pcie-bifurcation.1870298/page-19

[kravi4ka]

Can anyone,please, point me to the right size/type of Optane drive for the Euphony NUC7I7DNBE setup? I would hate to buy something that will not work, a shop in the EU would be wonderful. TIA

[Chopin75]

5 hours ago, austinpop said:

Mindful of @bobfa's point that these general topics are OT in the Shootout thread, I'm posting this here.

 

 

If there is one "new" finding in all of this recent progress, it is that combined with all the previously known optimizations of clean power, separate CPU power, linear supplies, better clocks, etc., running with a very powerful CPU like the i7-8700K provides a level of dynamism and clarity we have not heard from the low-powered CPU approach of the likes of the ZENith SE, the sMS-200ultra, and even the NUCs.

 

However the key thing here is that we operate these massively powerful CPUs at very low utilization - as we see with the non-upsampling use cases of Roon or Stylus.

 

Once you add upsampling into the mix, all bets are off, as it remains to be seen what the effect of that energy consumption due to high CPU utilization will be on:

1. the ability of fanless cases to cope with the added heat
2. the effect CPUs running at very high temps on SQ.

I had a chat with Nuno @nvitorino about this at Munich. I asked him if he would consider incorporating our findings in his server line - i.e. something like a ZENith or Statement containing powerful i7 and i9 cores. He said it was OK for us hobbyists to go this route, as we have the skill and knowledge to control our workloads and ensure we ran at very low utilizations. As a commercial vendor, he cannot control what people do with the box, and his concern is that people who try to do heavy DSP and upsampling may not get the SQ benefit - indeed, they may get the opposite effect. Also, this would put a strain on the PSUs, since they would have to design the PSUs for the heaviest case, which would require far more powerful and expensive PSUs than are currently in those boxes.

 

So this is a valid concern.

 

I think folks in the upsampling HQPlayer space need to run their own experiments and see where the sweet spot is. I would not be surprised if the answer is not standalone, but a distributed configuration: a separate powerful server for upsampling, and an endpoint, perhaps still using a powerful i7 or i9 core, but running NAA at very low CPU utilization. Please note - this is pure conjecture on my part, but it's an educated guess on what I and Roy have experienced thus far with non-upsampled Roon and Stylus.

I have been running upsampling/PCM —> DSD as well with high power CPU  Ryzen with fanless HDPlex, no issue  with overheat. Is SQ beneficial? Definitely much better than using my 2011 macbook with intel duel core chip. Though currently I don’t feel a need to upsample

[bobfa]

8 minutes ago, kravi4ka said:

Can anyone,please, point me to the right size/type of Optane drive for the Euphony NUC7I7DNBE setup? I would hate to buy something that will not work, a shop in the EU would be wonderful. TIA

Look for this in your country:

 

Intel Optane Memory M.2 2280 32GB PCIe NVMe 3.0 x 2 (MEMPEK1W032GAXT)

USA Amazon link:

https://www.amazon.com/gp/product/B06XSXX3NS/ref=ppx_yo_dt_b_asin_title_o04_s00?ie=UTF8&psc=1

[bobfa]

1 hour ago, lateboomer said:

Another thing is for single PC setup running Euphony Stylus, if os should be run in Optane card then where should music files be stored in optimized way?  Is it ssd or hdd? What is the type. I tried to skim through the index but cannot find the info. 

I have been playing with an external HDD (spinning) USB 3.0 drive with an LPS.  I have also been using a NAS. 

[Chopin75]

1 hour ago, lateboomer said:

Another thing is for single PC setup running Euphony Stylus, if os should be run in Optane card then where should music files be stored in optimized way?  Is it ssd or hdd? What is the type. I tried to skim through the index but cannot find the info. 

Not sure if Optane drive would be more beneficial, it runs fine with regular SATA drive I use a Samsung Evo. Any external drive may introduce moe noise so best is to keep music in the actual drive that has the euphony OS, or u load music from netwORK.  Euphony recommends keeping the music on euphony drive itself. And u can load music to ecache then detach the external drive afterwards. I realize that taking out the external drive helps improve the sound quite a bit. The sound still is affected even if u do not mount the drive onto euphony .

[lateboomer]

1 hour ago, seeteeyou said:

Assuming that RAM drives should be the very best for music storage in terms of latency, Optane (occupying processor-connected PCIe lanes) seemed to come pretty close according to Rajiv

 

https://audiophilestyle.com/forums/topic/30376-a-novel-way-to-massively-improve-the-sq-of-computer-audio-streaming/page/462/?tab=comments#comment-900522

 

Some folks actually managed to edit the BIOS and PCIe bifurcation was working afterwards, then Optane could be connected to the processor of an ITX motherboard directly

 

https://hardforum.com/threads/pcie-bifurcation.1870298/page-16

 

A few links to some risers, in theory we could have up to 4 processor-connected PCIe slots if there were a (fanless) case that's gonna fit

 

https://peine-braun.net/shop/

http://www.ameri-rack.com/ARC2-PELY423-C7_m.html

https://hardforum.com/threads/pcie-bifurcation.1870298/page-19

Thank you to all of you for the music storage info. I think the index should have separate section for music file storage. My case is I use this USB to sata adapter to connect to Samsung evo 850 ssd:

https://www.newegg.com/startech-usb312sat3-usb-to-sata/p/N82E16812400832

 

The ssd is partitioned to one partition for Daphile player and another one partition for music files.  I have created an USB cheater plug to supply 5v power to the adapter for the ssd and finally the USB adapter with the cheater plug is connected to the input of Startech 4-port usb 3.1 to pcie card. Generally I am OK with the sound that I get. I am still in a stage of making sense about these AL and Euphony approach and I am glad to learn from you guys. 

 

In side note,  the USB music also output from another independent port of this Startech 4-port USB to pcie card to USB dac. I supply external good smps to this pcie card.