Shootout at the Linux Corral: AudioLinux vs Euphony

[cat6man]

3 hours ago, austinpop said:

Some Euphony tunings:

1. Do not constrain the BIOS. Leave it wide open. By this I mean, turn on Turbo boost, HT, and don't run in low power mode.
   1. Then, play with max frequency in Advanced settings, starting at the max turbo freq and stepping down in small steps until any trace of harshness is gone. 
   2. With a really good PSU, you may not need to do this. 
   3. This is a big contributor to transient accuracy.
2. Apply core isolation. I'm assuming you have HT on — if not, turn it on. Try something like this:
   1. 0-1 stylus 2-3 gstp 4-7
   2. If others have better suggestions for NUC7i7 (t's been a long time for me), please chime in.
3. Turn on caching (Use cache flag) and buffering (buffer before play 100%, and buffer album)
4. Before playback on Stylus (local music, not Tidal/Qobuz):
   1. Add tacks to queue
   2. Hit buffer queue to RAM button on the top
   3. Wait for all the track status buttons (little round circles with i) to turn blue.
5. Now hit play

thanks for the suggestions @austinpop.

 

item 1:

i have HT and Turbo turned off, so i'll look into that tonight.

as for harshness, right now there is NONE as the sound is too smoothed so that is the direction needed.

item 2:

tbd (not yet used)

 

items 3, 4, and 5:

already in use here.

 

anyone have experience with the effects of turning hyperthreading and turbo ON for audiolinux?

[austinpop]

1 hour ago, cat6man said:

thanks for the suggestions @austinpop.

 

item 1:

i have HT and Turbo turned off, so i'll look into that tonight.

as for harshness, right now there is NONE as the sound is too smoothed so that is the direction needed.

 

Well, think about it. With Turbo off, your frequency is capped at 1.9GHz - the base frequency. With Turbo On, the frequency can go up to the peak of 4.2GHz (see https://ark.intel.com/content/www/us/en/ark/products/124968/intel-core-i7-8650u-processor-8m-cache-up-to-4-20-ghz.html). This region from base to peak is where the potential for better dynamics and transient coherence occur, but your PSU needs to be up to the task. Many PSUs will devolve into harshness in this range. The Max Freq control gives you a dial to find the operating point between 1.9 and 4.2 just at the onset of harshness (over the Turbo Off/Base case), while giving better dynamics and transients.

[cat6man]

2 hours ago, austinpop said:

 

Well, think about it. With Turbo off, your frequency is capped at 1.9GHz - the base frequency. With Turbo On, the frequency can go up to the peak of 4.2GHz (see https://ark.intel.com/content/www/us/en/ark/products/124968/intel-core-i7-8650u-processor-8m-cache-up-to-4-20-ghz.html). This region from base to peak is where the potential for better dynamics and transient coherence occur, but your PSU needs to be up to the task. Many PSUs will devolve into harshness in this range. The Max Freq control gives you a dial to find the operating point between 1.9 and 4.2 just at the onset of harshness (over the Turbo Off/Base case), while giving better dynamics and transients.

 

well, there is a little more detail with turbo off and hyperthreading on............no harshness though, but also not the amount of detail i hear with AL/squeezelite................no need to turn down the max frequency as it didn't get bright or harsh (hdplex300 psu)

[Lukasluis]

Exactly the same reasons why I switched to Windows. No amount of tuning cured my feeling that instruments are overly fat but lacked the leading edges and timing is quite off. No harshness even with 4.5Ghz on my Intel 8i7 wide open and all the tunings in place but I don't feel the excitement nor the emotion of the singers with Euphony. My license is unused for more than 4 months now. I am thinking of going 12 cores to see if things change but at the moment, I'm very happy with Windows. 

[austinpop]

Ok @cat6man. Just trying to help. Clearly my point is not getting through.

 

Good luck with with your experiments.

[cat6man]

1 hour ago, austinpop said:

Ok @cat6man. Just trying to help. Clearly my point is not getting through.

 

Good luck with with your experiments.

 

not sure what you mean.

 

i very much appreciate your recommendations, because you (and a few others) have consistently made useful observations that have aligned with my own with regard to SQ.

therefore, i was very interested in trying out your suggestions, which i think i have done now.

 

i am surprised, to some extent, that my results so far have been different from yours.

 

however, i'm not clear what exactly is the point you're making that "is not getting through"

please help me, what am i missing here?

 

 

[Lukasluis]

7 minutes ago, cat6man said:

 

not sure what you mean.

 

i very much appreciate your recommendations, because you (and a few others) have consistently made useful observations that have aligned with my own with regard to SQ.

therefore, i was very interested in trying out your suggestions, which i think i have done now.

 

i am surprised, to some extent, that my results so far have been different from yours.

 

however, i'm not clear what exactly is the point you're making that "is not getting through"

please help me, what am i missing here?

 

 

It looks like the Turbo Off. Did you try with HT On and Turbo On? 

[austinpop]

10 minutes ago, cat6man said:

 

not sure what you mean.

 

i very much appreciate your recommendations, because you (and a few others) have consistently made useful observations that have aligned with my own with regard to SQ.

therefore, i was very interested in trying out your suggestions, which i think i have done now.

 

i am surprised, to some extent, that my results so far have been different from yours.

 

however, i'm not clear what exactly is the point you're making that "is not getting through"

please help me, what am i missing here?

 

 

 

My key point was to turn turbo on, and low-powered mode off, then tune the max frequency. My own finding was that it was in this mode that the LPS quality and tendency to harshness can manifest. Hence the tuning process of adjusting max frequency.

 

Did you turn turbo on and try this? I'm not trying to convince you one way or another. Just want to make sure you're aware of my path. If you're unwilling to try it, that's your call.

[cat6man]

11 hours ago, austinpop said:

 

My key point was to turn turbo on, and low-powered mode off, then tune the max frequency. My own finding was that it was in this mode that the LPS quality and tendency to harshness can manifest. Hence the tuning process of adjusting max frequency.

 

Did you turn turbo on and try this? I'm not trying to convince you one way or another. Just want to make sure you're aware of my path. If you're unwilling to try it, that's your call.

 

that explains the miscommunication

that was a typo in my post, as i had both HT and Turbo on, and low power off.......i can see why i caused confusion

 

i knew what i meant to write and was responding based on that......oops

 

and thanks for the suggestions, i really was following them..........something must be different here.

as we all know, everything matters!

 

thanks again and sorry for the incorrect post above

 

 

[austinpop]

28 minutes ago, cat6man said:

 

that explains the miscommunication

that was a typo in my post, as i had both HT and Turbo on, and low power off.......i can see why i caused confusion

 

i knew what i meant to write and was responding based on that......oops

 

and thanks for the suggestions, i really was following them..........something must be different here.

as we all know, everything matters!

 

thanks again and sorry for the incorrect post above

 

 

Ah, that makes more sense.

 

Your observations are interesting. It sounds like the tunings I suggested did help, but not enough to change the outcome.

 

Do note that it’s been quite some time since I tried AL, so it’s very possible that that the improvements in AL have overtaken Euphony. 

[Lukasluis]

I too have tried Audiolinux and the best I experienced is euphony as a server and Audiolinux as an Endpoint. But for me, Windows makes me enjoy the music more. I can't explain why. I'm using Windows Server 2016 without any optimizations on a NUC 8I7 Akasa Fanleass case with a 8kg 19 volt Linear power supply. Perhaps the NUC is this good? I hope anyone here can chip in. I wonder if it is because of the SOC?

 

[AnotherSpin]

2 hours ago, austinpop said:

Ah, that makes more sense.

 

Your observations are interesting. It sounds like the tunings I suggested did help, but not enough to change the outcome.

 

Do note that it’s been quite some time since I tried AL, so it’s very possible that that the improvements in AL have overtaken Euphony. 

 

Once again I made a comparison between AL and Euphony endpoints some time ago. While Euphony gives full bodied, natural and free flowing musical sound, AL gives one which is lifeless, dry and excessively sharp in comparison. It may depend on hardware and other settings, of course.

[cat6man]

49 minutes ago, AnotherSpin said:

 

Once again I made a comparison between AL and Euphony endpoints some time ago. While Euphony gives full bodied, natural and free flowing musical sound, AL gives one which is lifeless, dry and excessively sharp in comparison. It may depend on hardware and other settings, of course.

 

i don't know how to interpret this, so could you please add some additional context?

comparison of endpoints for what mode, on what h/w?  can you describe the servers and how you configured things?

is this for lms/squeelite?

[AnotherSpin]

8 hours ago, cat6man said:

 

i don't know how to interpret this, so could you please add some additional context?

comparison of endpoints for what mode, on what h/w?  can you describe the servers and how you configured things?

is this for lms/squeelite?

 

HQP Desktop on mac mini with either AL or Euphony flash images on NUC 7i7. BIOS configured for max performance, NUC powered by Paul Hynes LPSU. I would confirm I prefer earlier AL builds as well, still, to my ears the sound is not as natural and effortless, as with Euphony. It may seem to  sound slightly more revealing with AL in a fast listen, however fatiguing after couple of hours or days.

[cat6man]

26 minutes ago, AnotherSpin said:

 

HQP Desktop on mac mini with either AL or Euphony flash images on NUC 7i7. BIOS configured for max performance, NUC powered by Paul Hynes LPSU. I would confirm I prefer earlier AL builds as well, still, to my ears the sound is not as natural and effortless, as with Euphony. It may seem to  sound slightly more revealing with AL in a fast listen, however fatiguing after couple of hours or days.

 

man, there are so many variables at play here.

i used hqp for a couple of years, with no signal processing at all, just benefiting from the separation of rendering and the NAA.  however, when i got a slimrun USB, i could run the NUC as the NAA (instead of a rendu), and then i found that with large buffers squeezelite sounded better than bit-exact hqp.

 

if you are running hqp with signal processing or upsampling or dither, you certainly have a different animal.

 

as for me, today i moved my LMS server off of the 7i7 NUC to a separate pc in the basement that runs ubuntu and also serves as my sageTV DVR and video server.................big big improvement (soundstage, detail) running the 7i7 as squeezelite only instead of running both LMS and squeezelite (i.e. 2 box solution instread of 1 box)....still running 7i7 with hyperthreading and turbo turned on, low power off in bios.

 

my networking was pretty well tweaked with a separate ubiquiti edgerouterX isolating the audio stuff from the rest of the house networking, then an etherRegen, all powered by some combinations of hdplex300, lps1.2, etc but i'll need to reconfigure things to optimize the new configuration.

 

i know euphony has a squeezelite end point option, and i'll try to compare that with AL.

is there any other euphony configuration as an end point that i should consider?

 

 

[AnotherSpin]

2 hours ago, cat6man said:

 

man, there are so many variables at play here.

i used hqp for a couple of years, with no signal processing at all, just benefiting from the separation of rendering and the NAA.  however, when i got a slimrun USB, i could run the NUC as the NAA (instead of a rendu), and then i found that with large buffers squeezelite sounded better than bit-exact hqp.

 

if you are running hqp with signal processing or upsampling or dither, you certainly have a different animal.

 

as for me, today i moved my LMS server off of the 7i7 NUC to a separate pc in the basement that runs ubuntu and also serves as my sageTV DVR and video server.................big big improvement (soundstage, detail) running the 7i7 as squeezelite only instead of running both LMS and squeezelite (i.e. 2 box solution instread of 1 box)....still running 7i7 with hyperthreading and turbo turned on, low power off in bios.

 

my networking was pretty well tweaked with a separate ubiquiti edgerouterX isolating the audio stuff from the rest of the house networking, then an etherRegen, all powered by some combinations of hdplex300, lps1.2, etc but i'll need to reconfigure things to optimize the new configuration.

 

i know euphony has a squeezelite end point option, and i'll try to compare that with AL.

is there any other euphony configuration as an end point that i should consider?

 

 

 

Yes, variables are plenty. Personal preferences are various as well, I  am looking for less disturbing, "analogue" type sound which may resemble that one from vinyl days. I use max power setting in BIOS as well. Did you try Miska's NAA for endpoint? All three (Miska's, AL and Euphony ) have their own characteristics. 

[AnotherSpin]

11 hours ago, cat6man said:

 

man, there are so many variables at play here.

i used hqp for a couple of years, with no signal processing at all, just benefiting from the separation of rendering and the NAA.  however, when i got a slimrun USB, i could run the NUC as the NAA (instead of a rendu), and then i found that with large buffers squeezelite sounded better than bit-exact hqp.

 

if you are running hqp with signal processing or upsampling or dither, you certainly have a different animal.

 

as for me, today i moved my LMS server off of the 7i7 NUC to a separate pc in the basement that runs ubuntu and also serves as my sageTV DVR and video server.................big big improvement (soundstage, detail) running the 7i7 as squeezelite only instead of running both LMS and squeezelite (i.e. 2 box solution instread of 1 box)....still running 7i7 with hyperthreading and turbo turned on, low power off in bios.

 

my networking was pretty well tweaked with a separate ubiquiti edgerouterX isolating the audio stuff from the rest of the house networking, then an etherRegen, all powered by some combinations of hdplex300, lps1.2, etc but i'll need to reconfigure things to optimize the new configuration.

 

i know euphony has a squeezelite end point option, and i'll try to compare that with AL.

is there any other euphony configuration as an end point that i should consider?

 

 

 

Your enthusiasm towards AL has been quite infectious 🙂 I flushed AL 1.4 to usb stick again and going to give it a run for couple of days. Listening to Scarlatti sonatas right now, piano sounds wonderful. Did you change any settings besides Priority set to extreme and enabling Ramroot?

[austinpop]

13 hours ago, cat6man said:

 

man, there are so many variables at play here.

i used hqp for a couple of years, with no signal processing at all, just benefiting from the separation of rendering and the NAA.  however, when i got a slimrun USB, i could run the NUC as the NAA (instead of a rendu), and then i found that with large buffers squeezelite sounded better than bit-exact hqp.

 

if you are running hqp with signal processing or upsampling or dither, you certainly have a different animal.

 

as for me, today i moved my LMS server off of the 7i7 NUC to a separate pc in the basement that runs ubuntu and also serves as my sageTV DVR and video server.................big big improvement (soundstage, detail) running the 7i7 as squeezelite only instead of running both LMS and squeezelite (i.e. 2 box solution instread of 1 box)....still running 7i7 with hyperthreading and turbo turned on, low power off in bios.

 

my networking was pretty well tweaked with a separate ubiquiti edgerouterX isolating the audio stuff from the rest of the house networking, then an etherRegen, all powered by some combinations of hdplex300, lps1.2, etc but i'll need to reconfigure things to optimize the new configuration.

 

i know euphony has a squeezelite end point option, and i'll try to compare that with AL.

is there any other euphony configuration as an end point that i should consider?

 

 

Does your NUC7i7 endpoint have 8GB or more RAM? You can try running SL with large buffers and see what you think.

 

On Euphony, just go into advanced settings, and enter this in the SL parameters field:

 

-b 2097152:2097152 -a 52428800:4::

On AL, last I did this more than a year ago, I had to edit the /etc/squeezelite.conf file. 
 

[cat6man]

3 hours ago, austinpop said:

Does your NUC7i7 endpoint have 8GB or more RAM? You can try running SL with large buffers and see what you think.

 

On Euphony, just go into advanced settings, and enter this in the SL parameters field:

 

-b 2097152:2097152 -a 52428800:4::

On AL, last I did this more than a year ago, I had to edit the /etc/squeezelite.conf file. 
 

 

yes, i have 8Gb of Acer RAM.

i've been running squeezelite (AL in ramroot) with the large buffers as my new reference (made possible by 50' slimrunUSB since my NUC does not co-habit with my stereo) per your old posts Rajiv.  it was a revelation and very pleasant surprise.

 

i plan to compare that with SL under euphony next, using the same buffer parameters.

if i understand correctly, this eliminates the pre-buffering of music that is possible with euphony as a server?

 

i also want to reconfigure my etherRegen.  now that i'm no longer using optical to feed the opticalRendu, i can put that to good use feeding my endpoint NUC.

 

thanks

[austinpop]

1 hour ago, cat6man said:

i plan to compare that with SL under euphony next, using the same buffer parameters.

if i understand correctly, this eliminates the pre-buffering of music that is possible with euphony as a server?

 

 

Yes, pre-buffering is only available with Stylus, and in this case you're using LMS. Still, the SL buffers do have a very good effect as you discovered. I assume you're using ramroot on Euphony too, of course.

[dminches]

On 8/28/2020 at 1:26 PM, austinpop said:

Some Euphony tunings:

1. Do not constrain the BIOS. Leave it wide open. By this I mean, turn on Turbo boost, HT, and don't run in low power mode.
   1. Then, play with max frequency in Advanced settings, starting at the max turbo freq and stepping down in small steps until any trace of harshness is gone. 
   2. With a really good PSU, you may not need to do this. 
   3. This is a big contributor to transient accuracy.

 

This is very helpful to know.  I am so used to the Windows Server / Audiophile Optimizer environment that I have always had Turbo and Hyperthreading turned off.

 

Can you tell me how things like c-states should be set?

 

I am going to listen with turbo on tonight.  My max frequency is boosted from 3.5 to 4.5 (Intel XEON E-2146G).  I have a Sean Jacobs DC4 so I am hoping I don't have to turn it down.  We shall see.

 

[austinpop]

18 minutes ago, dminches said:

Can you tell me how things like c-states should be set?

 

That's a good question, as there are advanced settings in some BIOSes for both C-States and P-States. I have not dipped down to this level to tweak, as it seems to be fraught with risk of shooting oneself in the foot! Maybe there are a few drops of SQ to be squeezed out of these settings, but it's not for the faint of heart, or those without a deep understanding of processor architecture.

 

The key thing to remember with Euphony is that the scaling governor (https://wiki.archlinux.org/index.php/CPU_frequency_scaling) is set to "Performance", so essentially you're running at the max frequency setting. You con confirm this by looking at the CPU monitor on the web UI. This monitor will show CPU frequency if either the Min or Max frequency fields are populated. If you want to see the current frequency without constraining, just set the Min to a low value like 800MHz. This will trigger the frequency display in the CPU monitor. 

[dminches]

20 minutes ago, austinpop said:

 

That's a good question, as there are advanced settings in some BIOSes for both C-States and P-States. I have not dipped down to this level to tweak, as it seems to be fraught with risk of shooting oneself in the foot! Maybe there are a few drops of SQ to be squeezed out of these settings, but it's not for the faint of heart, or those without a deep understanding of processor architecture.

 

The key thing to remember with Euphony is that the scaling governor (https://wiki.archlinux.org/index.php/CPU_frequency_scaling) is set to "Performance", so essentially you're running at the max frequency setting. You con confirm this by looking at the CPU monitor on the web UI. This monitor will show CPU frequency if either the Min or Max frequency fields are populated. If you want to see the current frequency without constraining, just set the Min to a low value like 800MHz. This will trigger the frequency display in the CPU monitor. 

 

I let my system warm up for 30 minutes and started listening to my test playlist.  I can hear a very positive difference after about 1 minutes with Turbo and hyperthreading on.  The difference isn’t subtle.  Much more relaxed presentation.  More detail but no harshness (thanks Sean).

 

I am glad I read your post.

 

Where is the CPU frequency monitor?  I see the monitor for temps and utilization but not frequency. 

 

NM.  I set the min to a high number and now it shows up at 4.3 GHz.

 

[austinpop]

19 minutes ago, dminches said:

 

I let my system warm up for 30 minutes and started listening to my test playlist.  I can hear a very positive difference after about 1 minutes with Turbo and hyperthreading on.  The difference isn’t subtle.  Much more relaxed presentation.  More detail but no harshness (thanks Sean).

 

I am glad I read your post.

 

Excellent. Always delighted to spread the joy around!

 

19 minutes ago, dminches said:

Where is the CPU frequency monitor?  I see the monitor for temps and utilization but not frequency. 

 

NM.  I set the min to a high number and now it shows up at 4.3 GHz.

 

 

I think you want to set the Min frequency to a low number. I've always used 800 MHz. This is only to trigger the frequency monitor. The OS never actually allows the CPU down to this frequency, as the performance governor effectively pegs the CPU to the max frequency.

 

For those who ask: but doesn't that defeat the purpose of energy efficiency and low noise from low power? No, it's just a different approach and hinges on the extremely low CPU utilization of bit-perfect playback. When a logical CPU is >99% idle, it's spending most of its time suspended, blocked on a HALT instruction. This is where the C-States come into play, but without getting into the weeds, the idea is that it is the idleness of the system that keeps it in a low-power, low-energy consuming state. And yet, when the CPU actually runs, it runs at high-frequency, which seems to deliver the maximum dynamics and transient response.

 

This entire premise falls apart on workloads like HQPlayer doing PCM to DSD upsampling, that have a heavy CPU load. This use case requires a different optimization, but I've not pursued it enough to have any expertise.

[dminches]

11 minutes ago, austinpop said:

 

I think you want to set the Min frequency to a low number. I've always used 800 MHz. This is only to trigger the frequency monitor. The OS never actually allows the CPU down to this frequency, as the performance governor effectively pegs the CPU to the max frequency.

 

 

Oh!