Euphony OS w/Stylus player setup and issues thread

[austinpop]

2 minutes ago, Johnseye said:

 

Yup I know. I could shift 2 SR7 rails in a single server environment. Will have to double check on the amperage of the second rail though. That could be an issue. Paul upgraded a component but not sure if I'm pulling a full 6A.

 

I did just get the HQPlayer NAA working on Euphony so I'm in business. Roon with HQPlayer on Euphony for now.  Will try that with Stylus and HQPlayer next. 

 

Yup I figured you were well aware of that! If you can get 2 SR7 rails deployed, it will be very interesting to see what you hear wrt harshness.

[bobfa]

46 minutes ago, austinpop said:

 

Yes, unfortunately this is a known issue with the NUC7i7. I ran into this a while back and discussed it with Željko. Since I could get the frequency changes to "stick" with AL, we hypothesized that it could be an issue with the older kernel that Euphony uses. In fact, Željko even went ahead and built me a prototype with the latest kernel, and the problem is fixed with the new kernel.

 

However... SQ degraded significantly with the new kernel. This is one of the dangers of this world. Euphony takes a very cautious approach to kernel upgrades, and they do careful listening tests before updating kernels and packages. Anyone using AL should be very careful before doing a kernel upgrade.

 

Anyway, for the time being, this does mean that frequency control is not working on the NUC7i7.

i was going to talk to Zeliko about my NUC and CPU frequency.  Guess I do not have to.

 

i need to get some notes at the top of this thread.

[mozes]

5 hours ago, RickyV said:

This configuration would be only beneficial for local files, right? Not tidal play

Actually it applies to both, local and Tidal

[Nenon]

@Johnseye

What CPU are you using? Also, are you using the ASRock Z390 Phantom Gaming-ITX/ac motherboard?

It would be really good to hear your report with a single server solution / Euphony powered by SR7 rails. 

[austinpop]

On the subject of harshness - there is another consideration when it comes to HQPlayer workloads. Unlike the non-upsampling Roon and Stylus workloads, where CPU utilization is < 5%, HQPlayer workloads run at much higher CPU utilization. You guys can tell me what you typically see.

 

Why does this matter? Given that Euphony pegs frequency with the CPU governor set to "performance," what difference does CPU utilization make on energy consumption and noise? I've been digging into this for my own edification. If you're not an OS geek, you can probably skip this!

 

From my understanding, the answer lies in the concept of C states. Here's a decent intro to C States: https://www.dell.com/support/article/us/en/04/qna41893/what-is-the-c-state?lang=en In a nutshell, when a <HW thread|core|CPU> becomes idle - i.e. the OS has no work (runnable thread) to dispatch - it executes a HALT instruction, that puts it into a lower power C state (C1). One of the key things about this state is that the clock is turned off. From this point, there is a whole plethora of deeper sleep states that progressively lower energy use. Here's a nice picture of what each C state entails:

 

How does this relate to our harshness discussion? Even with frequency pegged to the max, when a workload is running at a very low CPU utilization, these CPUs spend a lot of time with the core clocks off, so this is why core temps don't rise that much, and harshness stays under control. If we had root access in Euphony, we could verify this hypothesis by running turbostat with different workloads. This would tell you the percent of time the CPUs spends in C1 through Cn state.

 

For those of you who have used AL, this also explains the EXTREME and EXTREME2 modes in that OS. EXTREME boots the kernel with a flag setting: intel_idle.max_cstate=0. In the context of the preceding para, you can now see this disables cores from ever leaving C0 state. In EXTREME2, there is an additional flag: idle=poll. This disables even the execution of HALT, and instead causes idle HW threads to run a "no op" idle loop. Theoretically, these flags reduce the latency to dispatch new work, but they have a severe impact on energy - and by association, noise.

 

Hope this helps those interested in a little deeper look.

[ted_b]

Rajiv,

Thanks for this!  Although this stuff is absurdly over my head technically, I still find it fascinating.  The unearthing of previously unknown or poorly understood causes of noise are fascinating to me.  We begin this computer audio journey with so little real information, enjoy our sq for awhile, make some changes and voila!  The hint of until-now unheard magic flows from our speakers, and we are hooked.  Confident that more gremlins (often unmeasurable at the time) eliminated means closer to the audio truth.  And these discoveries seem to come in ebbs and flows, with right now a very opportune time...thanks to the likes of you, Romaz, Larry, Bob and all the clever other early adopters (not to mention the developers themselves).

 

I sit on the sidelines, waiting my turn, as I hope and pray my audio room gets built before I get too much older. 

[bobfa]

So there may be two things conspiring here that I did not see.   I was always running AL in Extreme because I thought that it was the right thing to do.  So with Euphony we do not disable those states.  I know that there is more to this but the data is interesting.

 

The other note that Euphony is using a different kernel version than AL is also fodder for the sound mill.

 

 

[Chopin75]

21 minutes ago, austinpop said:

On the subject of harshness - there is another consideration when it comes to HQPlayer workloads. Unlike the non-upsampling Roon and Stylus workloads, where CPU utilization is < 5%, HQPlayer workloads run at much higher CPU utilization. You guys can tell me what you typically see.

 

Why does this matter? Given that Euphony pegs frequency with the CPU governor set to "performance," what difference does CPU utilization make on energy consumption and noise? I've been digging into this for my own edification. If you're not an OS geek, you can probably skip this!

 

From my understanding, the answer lies in the concept of C states. Here's a decent intro to C States: https://www.dell.com/support/article/us/en/04/qna41893/what-is-the-c-state?lang=en In a nutshell, when a <HW thread|core|CPU> becomes idle - i.e. the OS has no work (runnable thread) to dispatch - it executes a HALT instruction, that puts it into a lower power C state (C1). One of the key things about this state is that the clock is turned off. From this point, there is a whole plethora of deeper sleep states that progressively lower energy use. Here's a nice picture of what each C state entails:

 

How does this relate to our harshness discussion? Even with frequency pegged to the max, when a workload is running at a very low CPU utilization, these CPUs spend a lot of time with the core clocks off, so this is why core temps don't rise that much, and harshness stays under control. If we had root access in Euphony, we could verify this hypothesis by running turbostat with different workloads. This would tell you the percent of time the CPUs spends in C1 through Cn state.

 

For those of you who have used AL, this also explains the EXTREME and EXTREME2 modes in that OS. EXTREME boots the kernel with a flag setting: intel_idle.max_cstate=0. In the context of the preceding para, you can now see this disables cores from ever leaving C0 state. In EXTREME2, there is an additional flag: idle=poll. This disables even the execution of HALT, and instead causes idle HW threads to run a "no op" idle loop. Theoretically, these flags reduce the latency to dispatch new work, but they have a severe impact on energy - and by association, noise.

 

Hope this helps those interested in a little deeper look.

Not entirely sure what you mean with C states, but my CPU is usually running  < 5% for all cores even with HQplayer upsample from 16/44hx ---> 24/352  but occasionally jumps top to 10%. (From Euphony OS, it shows I have 16 CPU, but only about 5-7 are used at anytime) Both Stylus vs HQplayer embedded are similar in CPU usage.  I use AMD Ryzen 2700x 8 core.   I actually prefer the STylus upsample right now as the HQplayer, while smoothening things a lot, seems to do so too much so the natural instrumental color is altered but I have not had chance/time to adjust the different filters yet. THe only time CPU goes up to >20% is when I use HQplayer upsample to DSD256.  Hope this helps. 

[Chopin75]

7 hours ago, sergiocurvelo said:

 

Bob,

Two days  testing various configurations and I can assure you that Stylus + HQ Player is the best setting I've tested here. More natural textures, analog, energetic and lively sound.

As the server is near my DAC, I will no longer need the Ultra Rendu and I'll sell it. I'll connect the Curious USB cable in the DAC working along with ISO Regen.

 

Maybe Roon's annual subscription will not be renewed. Until November I will decide.
But I'm really enjoying it: Stylus+Tidal+HQ Player

 

Can you clarify what Stylus + HQplayer means? As far as I know, you can only play either Stylus or HQplayer embedded or NAA. When you play HQplayer embedded the Stylus is actually disabled and the HQplayer takes over. Do you really mean using Stylus as library management but HQplayer that actually plays the music files ?  

[Chopin75]

9 hours ago, Superdad said:

HELP!!

 

So I have wasted a bunch of hours this holiday weekend just trying to get Euphony booted and a connection to it.  They sure have some fix-in to do.

 

Setting aside my end objective configuration I'd like to audition it in, I'm just trying to get it going at all.  No problem creating a Euphony boot USB stick using Etcher (have used that and other utils many times to make bootable Linux sticks).  But despite Euphony's indication that they support UEFI booting on Macs, here is what happens once I boot to the image--same on 3 different Macs:

 

 

 

So then I figure I'd try booting this thumb drive on either my NUC or my Win10 laptop.  I can get either one to boot successfully--to the point where it says "EUPONY IS UP AND READY," please access on a web browser, etc.

 

Problem is, Euphony self-assigned itself an IP address--on a sub-net different from what my ADSL modem/router hands out.  On both machines it to 192.168.1.237.   But all addresses on my network are 192.168.8.xxx.

So of course attempting to access it from a browser does not work.

 

Suggestions anyone?

Thanks,

ALEX

 

P.S.  Regarding the Mac efivarfs/emergency shell issue, an hour ago I submitted a ticket, and now received the following reply:

 

"This is very peculiar issue with iMacs. You have to install gdisk command line utility.
http://macappstore.org/gptfdisk/
After that use gdisk /dev/disk3 (or whatever number is USB with Euphony on it) and then enter these
x
e
w
y
(Enter after every letter).
I'm writing from phone so I'll explain later what this is and why this problem occurs.
And use USB stick, not SD card because with them there is no solution at all."

 

That's nice I guess, but I replied that the issue is not just with iMac.  I at first tried two different Mac minis with same result.

What yrs are your Macs made ? Any Mac that has T2 Chip won't work at all with external USB boot drive or any other type of foreign non-mac OS reboot for that matter. It is the nice high security of Mac (you see, this bans anyone trying to reboot a mac into PC or Linux mode in case someone try hack in ???) ! All 2018 Macs whether macbook, macmini etc have T2 chip. 

[Superdad]

38 minutes ago, Chopin75 said:

What yrs are your Macs made ?

 

Late 2012 mini, Late 2014 iMac Retina.  Both boot fine from other Linux USB sticks.

[Chopin75]

3 minutes ago, Superdad said:

 

Late 2012 mini, Late 2014 iMac Retina.  Both boot fine from other Linux USB sticks.

Strange, but it may be due to V3 version of the Euphony. I no longer can boot from my macbook 2011 using the V3 version. There is a complex way to make it work for V3.  But the PC that you booted from seem to work. Probably should just try that

[Holzohr]

2 hours ago, Chopin75 said:

Can you clarify what Stylus + HQplayer means? As far as I know, you can only play either Stylus or HQplayer embedded or NAA. When you play HQplayer embedded the Stylus is actually disabled and the HQplayer takes over. Do you really mean using Stylus as library management but HQplayer that actually plays the music files ?  

 

You can use Stylus with the HQPe as playback engine. Please have a look at the Stylus settings.

[HeeBroG]

11 hours ago, mozes said:

I reached the same conclusion on my system even with the tX-USBexp. I settled on LMS-->StylusEP. This takes away the harsh edges on the highs. 

 

Hi mozes,

 

What machine are LMS and Stylus EP running on?

 

Geoff

[mozes]

25 minutes ago, HeeBroG said:

 

Hi mozes,

 

What machine are LMS and Stylus EP running on?

 

Geoff

Hi Geoff

Basically both are low powered machines.

I have Nimitra running LMS and a custom sCLK-EX PC  based on a Jetway NF596 motherboard with Celeron N3160 CPU running StylusEP.

[motberg]

10 hours ago, austinpop said:

I notice several reports of increased harshness, loss of smoothness, and the like associated with trying Euphony on a standalone server, whether with Stylus, Roon, Roon+StylusEP, HQPlayer, Roon+HQPlayer, and Stylus+HQPlayer.

 

I've written about this myself, and it's the reason I don't consider my system finished just yet. What going to a powerful standalone system, compared to NUCs and other endpoints, gives you is big, weighty, DYNAMIC sound. However, this puts a bigger burden on the quality of the PSU on the server. And limitations of PSU quality manifest as harshness.

 

Very few of us besides @romaz have SR7 rails powering our server's ATX and EPS inputs. In my system, with the HDPlex 400W LPS powering the ATX and EPS inputs, and with an SR4-5 powering the JCAT Net Card Femto, I still find lingering harshness to be a problem.

 

My solution, and something I suggest people try, and report back - is to tune the "Max frequency" in Expert settings.

 

For example, on my server that has the i7-8700T CPU, the base frequency is 2.4GHz, and the peak is 4.2GHz. I have set the BIOS to be wide open - i.e. I've enabled Hyperthreading and Turbo. With Min and Max frequencies unspecified, Euphony's frequency hovers between 3.9 and 4.0GHz. This behavior is expected because - as confirmed by Željko - Euphony is preconfigured to use the "performance" CPU governor, so frequency is effectively pegged to the max allowed.

 

In my system, I can almost completely tame the harshness by setting the Max to 2.4GHz (or 2400MHz). Based on trial and error, I find a minimal rise in harshness up to 3200MHz, and depending on the music, I can stand to go up to 3600MHz.

 

Why bother to skirt this harshness boundary? In a word - dynamics! That is the elixir that this config provides, and I think we all have different thresholds of what we're willing to pay for it.

 

So bottom line - to those experiencing harshness with Euphony - try tuning the Max frequency and see what you think. Please report back.

I wonder what would happen if you power the EPS with a separate LPS ....

I have my audio PC setup like that - ATX 20 pin with ATX LPS, and then a separate 10A 12V LPS for the EPS (Ghent cable)

[sergiocurvelo]

4 hours ago, Chopin75 said:

Do you really mean using Stylus as library management but HQplayer that actually plays the music files ?  

 

Exactly

[Johnseye]

8 hours ago, Chopin75 said:

Can you clarify what Stylus + HQplayer means? As far as I know, you can only play either Stylus or HQplayer embedded or NAA. When you play HQplayer embedded the Stylus is actually disabled and the HQplayer takes over. Do you really mean using Stylus as library management but HQplayer that actually plays the music files ?  

 

3 hours ago, sergiocurvelo said:

 

Exactly

 

In Euphony's settings, after you enable Stylus as your player, you should be able to select HQPlayer as your upsampler.  There's a limited selection of parameters provided.  You can also direct HQPlayer to an NAA in this mode.

 

Euphony uses HQPlayer Embedded.  I believe it's v4 but have not confirmed this.  There's no plan for them to use the Desktop version and it will require a separate license or else you're limited to 30 minute sessions, then need to restart the service.

[Johnseye]

9 hours ago, Nenon said:

@Johnseye

What CPU are you using? Also, are you using the ASRock Z390 Phantom Gaming-ITX/ac motherboard?

It would be really good to hear your report with a single server solution / Euphony powered by SR7 rails. 

 

It's not the Z390, it's a Z270 chipset.  The board was modified by SoTM with the sCLK-EX.  It has no trouble upsampling DSD to 512.  The difference between the 200 and 300 series chipsets are minimal so that's not a concern for me, and it can run Optane if I need that.

 

If I do this with Euphony and use both rails to power this board, I'll need to have Ghent build me a cable.  I have some more tinkering to do with a 2 box solution before that.  I was able to smooth out the edge by getting HQPlayer to point to the NAA.  What I need to do now is put some time into listening, then swap back and forth between Euphony and AL.

[bobfa]

8 minutes ago, Johnseye said:

 

It's not the Z390, it's a Z270 chipset.  The board was modified by SoTM with the sCLK-EX.  It has no trouble upsampling DSD to 512.  The difference between the 200 and 300 series chipsets are minimal so that's not a concern for me, and it can run Optane if I need that.

 

If I do this with Euphony and use both rails to power this board, I'll need to have Ghent build me a cable.  I have some more tinkering to do with a 2 box solution before that.  I was able to smooth out the edge by getting HQPlayer to point to the NAA.  What I need to do now is put some time into listening, then swap back and forth between Euphony and AL.

I am very interested to hear your results.  

 

I feel a bit queasy with everyone doing all of this testing.  While we are on an interesting quest, i hope we are not Don Quixote tilting at windmills. 

 

I feel that this method of disconnected communications about separate executions of our own thoughts, seems to be a detractor to progress. At the moment I cannot think of a better way to coordinate efforts for better effect using only the forum software.  IF we extend communications with Slack and maybe Trello we might create a way improve communications and progress.  

[vassils]

4 minutes ago, bobfa said:

I am very interested to hear your results.  

 

I feel a bit queasy with everyone doing all of this testing.  While we are on an interesting quest, i hope we are not Don Quixote tilting at windmills. 

 

I feel that this method of disconnected communications about separate executions of our own thoughts, seems to be a detractor to progress. At the moment I cannot think of a better way to coordinate efforts for better effect using only the forum software.  IF we extend communications with Slack and maybe Trello we might create a way improve communications and progress.  

 

Bob - I hear what you are saying, but with people on different schedules, different timezones, daily jobs, family, different priorities, etc. there is no easy way to do that. We can try to create a local Chicago group, though, and improve coordination and communication between us. And report back our findings here. 

[johnli]

hi all

fascinating discussion

 

I've finally bought and  installed euphony and will compare it serious with AL again

There was a lot of hiccups for installation, and changing to a new USB stick and new M2 SSD saved the day.

 

Anyway, there is a new update just released today. I forgot to read the content and then just pressed update.

 

Any ideas what the update is about?

 

Thanks

[Johnseye]

2 minutes ago, johnli said:

hi all

fascinating discussion

 

I've finally bought and  installed euphony and will compare it serious with AL again

There was a lot of hiccups for installation, and changing to a new USB stick and new M2 SSD saved the day.

 

Anyway, there is a new update just released today. I forgot to read the content and then just pressed update.

 

Any ideas what the update is about?

 

Thanks

 

Funny you should ask.  I just received this email from Željko...

 

Hi John,

Just wanted to inform you that I issued an update which fixes the issues you've mentioned:

- HQPlayer settings are always available  (you don't have to have some audio card present)

- Auto is added to selection

- If you have multiple cards on NAA you can select which one to use

 

Regards, 

Željko

[Dev]

12 hours ago, austinpop said:

 

Yes, unfortunately this is a known issue with the NUC7i7. I ran into this a while back and discussed it with Željko. Since I could get the frequency changes to "stick" with AL, we hypothesized that it could be an issue with the older kernel that Euphony uses. In fact, Željko even went ahead and built me a prototype with the latest kernel, and the problem is fixed with the new kernel.

 

However... SQ degraded significantly with the new kernel. This is one of the dangers of this world. Euphony takes a very cautious approach to kernel upgrades, and they do careful listening tests before updating kernels and packages. Anyone using AL should be very careful before doing a kernel upgrade.

 

Anyway, for the time being, this does mean that frequency control is not working on the NUC7i7.

 

Well, I revisited CPU frequency settings in Euphony on NUC7i7. Seems like it sticks to the frequency if the set frequency is less or equal to CPU base frequency. For NUC7i7, the base frequency 1.9Ghz. So anything 1.9Ghz and below sticks. Anything above that doesn't. Interestingly if I set it to 2.4Ghz, the frequency hops from 3.5-3.7Ghz (mostly at 3.7Ghz) but if I set it to 1.95Ghz, then it wanders till 4.1Ghz. 😊

[austinpop]

12 hours ago, Chopin75 said:

Not entirely sure what you mean with C states

 

Well, that's what my entire post starting from paragraph 3 was about!

 

12 hours ago, Chopin75 said:

but my CPU is usually running  < 5% for all cores even with HQplayer upsample from 16/44hx ---> 24/352  but occasionally jumps top to 10%. (From Euphony OS, it shows I have 16 CPU, but only about 5-7 are used at anytime) Both Stylus vs HQplayer embedded are similar in CPU usage.  I use AMD Ryzen 2700x 8 core.   I actually prefer the STylus upsample right now as the HQplayer, while smoothening things a lot, seems to do so too much so the natural instrumental color is altered but I have not had chance/time to adjust the different filters yet. THe only time CPU goes up to >20% is when I use HQplayer upsample to DSD256.  Hope this helps. 

 

Thanks, it's good to get these CPU utilization data points. My concern about CPU utilization was mostly aimed at the common use case of upsampling to DSD512.