Euphony OS w/Stylus player setup and issues thread

[edwardsean]

I'm using Euphony/Stylus installed onto Optane and RamRoot on Apacer RAM. I have both full song buffer and cache enabled. Since the tracks are being played from Optane and RAM does this mean that it doesn't matter where the song files are coming from.

 

I've been using an SD Card to avoid issues with SSD and Hard Drive. However, it is of course pretty slow, and my song files are upscaled PCM768KHz so are about 1GB each. 

 

Would I not lose any SQ using a faster drive?

 

[edwardsean]

So, if I'm following this correctly:

 

All the traditional SQ benefits of using a NAS instead of a directly attached drive, and the sonic issues of an SSD, no longer apply with Euphony buffering songs to RAM?

[edwardsean]

This is the message I got from Željko:

 

"I often hear that people avoid SSDs because they think SSD is noisy even when not being actively read but that probably varies from one SSD to the next and I don't really expect you will be able to hear the difference when playing from RAM unless your system is already ultra resolving."

 

I spent some time comparing an external SD Card and SSD. My system is very resolving, and I do think there is a slight difference. There did seem to be a–slightly–excited upper mid/highs which is consistent with increased noise. It also just sounds a bit more clear. However, as he said, playing from RAM, it was not a major difference. 

 

Since the audio file is not playing directly from SSD but RAM, we are talking about just the noise emission of having it plugged in and powered. 

[edwardsean]

Thanks Dave, 

There are a few things to consider here, one is that you're comparing a SDCard, SSD connected locally and the HD was connected over network. Did you try connecting a HD directly to the NUC and were the results the same?

 

Also, in your second post you mentioned you moved to RAM mode. I assume you mean RamRoot. However, in your tests were you caching to an Optane drive and buffering 100% song to RAM before playback. 

 

This is the use case that I'm trying to test. The question is: if audio files sound different played directly from SDCard, SSD, or HD, does this difference hold when the files are no longer being played directly but from cache/RAM?

[edwardsean]

Yeah, I surprised to find how much difference Apacer made. I only wish the 2666MHz sticks were readily available. It took some doing to track down used. Definitely worth it though. 

[edwardsean]

I have huge files myself, and with only 8GB of industrial grade RAM, I run out all the time. I have 32GB of regular consumer grade RAM, which I don't use. It was the sacrifice I made to use the Apacer RAM.

 

What high quality RAM are you running that you were able get 32Gb? 

[edwardsean]

Just a quick caveat on BIOS. I don't know if this is common knowledge but, I went into BIOS to check that I had everything setup for maximum performance, which I did.

 

While I was in there I saw that under cooling, my fan preference was set to "balance" (default). I switched it to "fanless," because I'm using an Akasa fanless case. So, I thought this would either make no difference, or possibly add a slight optimization by defeating fan polling. 

 

I started up Euphony and it was struggling with normal operation. It seems that setting the BIOS to "fanless" throttles performance. I've read that it disables Turbo and/or HyperThreading. That may be right even though both still show up as active. 

 

At any rate, I found it severely hampers Euphony, at least the way I'm running it. Going back into BIOS and setting fan to "balanced" and everything is right as rain again. 

[edwardsean]

Been tinkering with CPU Isolation with interesting results. The difference between the default and [0-1 stylus 2-3 gstp 4-7] was pretty subtle. So, I experimented further. 

 

I found Nenon's observations held true for me. Isolating stylus didn't seem to make a big difference, but isolating gstp yielded benefits. It seems the more cores you can give gstp the better.

 

I ended up going back and forth between [0-2 gstp 3-7] and [0-1 gstp 2-7]. The latter seemed to jump a threshold and provided a noticeably clearer, denser, and definite sound. Going under 6 cores for gstp produced a pleasing and airy but more diffuse, unfocused sound. On balance, gstp 3-7 was smoother where gstp 2-7 might get ever so slightly edgy. However, it's hard to say that this edge was noise or from pushing the system vs. a more revealing clarity. 

 

In the end [0-1 gstp 2-7] is worth whatever tradeoffs to my ears. 

 

This little test made a good case for me to build a full high power server. If the kind of sonic improvement I heard keeps gaining as you supply gstp with more cores and power it would be worth it. 

 

Can some users who have gone from a NUC (esp. 717DNBE) to full servers share their findings in terms of level of improvement? 

[edwardsean]

I'm continuing to play around with core isolation while I listen and make changes to my system. I've found it is just not a matter of giving gstp as many cores as you can. At least if you are only working with 8 cores or less. The more you devote to gstp the more the sound gets solid and defined. However, like with processing speed, it can come at the cost of bringing out more edge and unwanted brightness. I've backed off from 0-2 gstp 2-7 to 0-2 gstp 3-7. 

 

This is right for my current configuration, but of course this is all so system dependent. Calibrating the isolation on Euphony is interesting. As long as you have a highly resolving system the changes are immediately apparent. I've even found a difference between default and 0-7 stylus 0-7 gstp 0-7, the latter being more clear but less fluid. 

 

Some might think of the changes as subtle and unnecessary, but if you've come this far, having this bit of tuning is very satisfying. 

[edwardsean]

Cosmetic changes in case anyone is interested:

 

- Play que is now dark grey instead of light grey. 

- Track selection bar is now flat teal color

- Background album art is darker and less distracting

- Now playing cover art resolution has bumped up from 300px, though still not full quality. 

 

Anwar's notes above say:  "Use original cover for Playing now screen" 

Anyone know what this means? Original cover art always showed for me in the Playing now screen. 

 

 

[edwardsean]

I like the aesthetic changes, though I no longer think the cover art resolution was bumped up. I really do wish that the now playing album art would display at full resolution. It's part of the experience for me. 

 

I'm still listening for changes to sound. I do detect an increase in terms of clarity, refinement, imaging, and detail on my Dave. If this is the case, so far, I can't hear any sacrifices to fullness, smoothness or the weight of notes on balance. I still need to hear more. I know how much the developers care about preserving sound quality over updates. I think this is just pure gain and I don't think any users would want to go back as a matter of preference. 

 

What is certain for me is that performance has seemed to take quite a leap forward in my system. I'm running Euphony on a NUC7i7DNBE. Previously there were issues which I chalked up to the fact that I'm using files upscaled to 768KHz. My average filesize is around 1Gb. Euphony would take a bit of time loading them into Cache/Ram even from my USB3.0 SSD. Also, If I use skip, repeat, or move the playhead the sound would stop and I would get a loud burst of static and I would have to restart the app. 

 

With the update, the tracks load quickly and I can move around without Euphony struggling or breaking up into static. Fantastic. 

 

Thank you Euphony team!

[edwardsean]

Another new update went live. Just a few a minor fixes. 

[edwardsean]

Okay, I found out that "now playing" page can now use cover art stored in the folder and displays it at full resolution. Fantastic.

 

Thanks to Euphony! 

[edwardsean]

 A couple days ago I made a DC umbilical for the Sean Jacobs DC3 which powers my DAVE. I followed Nenon's suggestion of using Mundorf Silver/Gold. I'm posting about it here, because the stuff added such a nice smoothness and density to my sound I decided revisit my Euphony core isolation. 

 

I settled in at 0-2 gstp 3-7. Giving more cores to gstp seems to improve the sound but at the cost of some thinness and stridency. 

 

However, with the new DC umbilical I went up again to 0-1 gstp 2-7 with some nice results. So, I decided to go further than I had before and tried 0 gstp 1-7, giving the playback engine 7 cores. 

 

I don't know, if your system can handle it, the benefits to SQ are knocking me out. More dynamics, clarity, soundstage, definiteness, detail, and extension in the highs. gstp 2-7 also lost some 80KHz bass punch to gstp 3-7. gstp 1-7 retains low end as well. 

 

I've been watching temperatures, because 3 of the 4 physical cores have dropped in temp, but core 1 is now up and holding around 74 degrees. That's acceptable to me.

 

if you have a 8 core NUC can you test this and see what you think? Again, this is all system dependent, this wouldn't have worked a couple days ago. 

[edwardsean]

The first range of core numbers are for general processing of Euphony. Then comes stylus, and next Roon, HQP, etc. I don't use either so I forget the process names. However, once you hit apply, you'll get the screen that shows the processes and core assignments. Roon should be easy to identify. "gstp" is the actual playback processor and so, personally, I've found that giving it more cores is beneficial, up till the point it causes harshness. 

 

The image below is a good starting point. Along with some others, I haven't found isolating stylus did much and that devoting cores to gstp did a lot. So what I posted above (0 gstp 1-7) gives one core to everything including stylus and reserves 7 cores for gstp. This seems extreme to me even as I'm doing it so it may not work well in other systems.

 

Good luck and report back if you can!

 

 

 

 

[edwardsean]

Does anyone know if min frequency functions now?  I haven't found that it does.

[edwardsean]

I put my min frequency up to 4000000 and Euphony shows the frequency dipping down below to 3.8-3.9GHz all the time. 

 

I'm not sure setting the min frequency does anything. 

 

 

[edwardsean]

1 hour ago, TheAttorney said:

So good news, and now the better news: Whilst I broadly agree with your findings, I think I've found an even better combination than [0 gstp 1-7]. To my ears, [0-1 gstp 1-7] gives even more of the above SQ benefits, plus the further bonus of a significantly lower temperature for physical core 0. The temperatures are now more balanced across the cores, which isn't intuitive because it's the same physical core 0 in both cases. And virtual core 1 is now sharing gstp with everything else. Core Isolation moves in mysterious ways.

 

First of all, I know you must be a lawyer because you use the term "whilst" (grin). I do love the English language. 

 

Awesome! I'll try out 0-1 gstp 1-7 tonight. If I can improve SQ and lower temps that's a good day. Though you are quite right it is mysterious with so many system specific factors. I'm always at 3.9-4.2GHz so the results may vary. 

 

Isn't it a kick in head how core isolation affects SQ as much as it does? Also, I'm amazed that you can just adjust these parameters in real time and hear the changes. If you had to reboot each time I would constantly be questioning if I'm hearing the changes correctly. As it is it's just so handy to be able to assign cores whilst listening (I couldn't resist because of the assonance between 'whilst' and 'listening.' I do love the English language). 

[edwardsean]

I tried out 0-1 gstp 1-7, and the temps got displaced to the first two cores. So 1 went down and 2 went up; both are low 60s. This is welcome.

 

To my ears the sound suffered slightly. It's really subtle, but it sounded smaller and flatter to me. 0 gstp 1-7 still sounds a bit larger and more vivid to me. 

 

This could just be my system or psychology. This is complicated, however, my guess is sill that the more cores you dedicate to gstp the better. Thinking simply, sharing core 1 reduces isolation. 

[edwardsean]

Thanks for letting us know. I'm doing the update right now. Euphony is really putting out the updates. 

 

Please report if you hear any changes to sound.

 

Here is the latest changeling:

 

20201102
- WARNING for users on version less than 2020-10-05: TIDAL has changed authorization process and TIDAL will stop working sometimes mid December unless you update.
- Added processing of CUE playlist files
- Reorganization of Settings page
- Added some system info fields
- Added some Music DB stats
- Added some Summus specific UI changes
- Euphony launch page improvement: started audio service is displayed in addition to hostname
- Youtube playback fix
- Fixed adding multiple network drives
- Fixed Tidal expire problem
- Fixed Tidal authorization from native Android/iOS applications (they are updated - download the latest version)
- Fixed the problem when additional internal drives won't remount on startup
- Fixed radio playback mode

[edwardsean]

Upsampling to higher sample rates does not automatically mean better sound. Actually, it's the opposite for the basic filters that come with most apps. Unnecessary upsampling will only make the sound worse, and the further you get from the source frequency the more error you introduce.

 

The computationally heavy algorithms implemented in HQP and Mscaler are incredibly complex and are of a different nature. They don't only increase the sample rate, but attempt to reconstruct the original waveform. So, in their case the more samples you can give it the more accurate they become. 

 

Unfortunately, the readily available sampling filters (e.g., SoX) are not free/cheap ways to mimic the results of HQP/MSc. It's not their purpose/design. You want to use them only for simple corrections: 48KHz to 44.1KHz. 

[edwardsean]

As to loading songs to memory, no features have been removed as you know, just reorganized. 

 

The loading to RAM, "buffer before play," is in: Settings > Music service > Other music options.

 

You can load the tracks to a cache if you have a fast drive with "Use cache" in: Settings > Library > Miscellaneous. 

 

 

[edwardsean]

On 1/26/2021 at 2:48 PM, ASRMichael said:

I use this Apacer RAM for my NUC

 

https://www.mouser.co.uk/ProductDetail/908-D23.23240S.004

 

If i recall Nenon recommended it

 

Is this RAM non-ECC then?

 

I'm currently using 2 sticks of 4GB Apacer D21.23180S.001 from Nenon's group buy. However, I'd really like to get two sticks of 8GB RAM.

[edwardsean]

Does anyone have news on Euphony v.4 that they are allowed to share?

 

Is this release being developed as a step up in SQ, features, both?

 

Projected release date?

[edwardsean]

FYI: New update guys.