brucew

When I changed from CD’s to file-based playback 12 years ago, I fed the DAC with a Fidelized and tweaked laptop playing JRiver. It had pretty decent sound but I knew I should be able to do better and so went to microRendu for a low noise option. I used a microRendu and UA Ultracap LPS-1 for a few years with a remote fileserver, but always felt like it lacked the sense of timing and very low noise floor that I’d had with a CD transport-based system. So, in 2022 I switched to an Audiolab 6000CDT transport, started spinning discs again and was rewarded by what I’d been missing. After a year or so with that I wondered if I could get to that same sound with a file-based player without many boxes and 10x the price of my disc spinner, so started experimenting and using my recent experience with the CD transport as a reference. Most of these ideas are not novel but are spread across many threads, but those threads are often focused on upsampling and/or cost-no-object solutions. My budget requires me to set different goals, plus I wanted to put these together in one place. And I wanted a headless system, so no screen interfering with the acoustics. These are items that received my attention: Choice of processor Power Supply Local storage vs networked storage SSD vs HDD Fan vs fanless USB2 vs USB3 CPU frequency Status update frequency Processing & Volume control Hyperthreading & Turbo on/off For software, I tried JRiver Media Center and Euphony Stylus on a NUC, sometimes with the files in a separate NUC as server and sometimes as a separate drive inside the player. Choice of Processor If going with a player that doesn’t upsample or use Roon, processor power doesn’t seem to be an issue. So, I reasoned that a Core i3 should be sufficient for anything 5th generation or newer, certainly no more than a Core i5. The greater the power capacity the item is designed for, the more electrical noise is likely to be an issue, effecting the noise floor. It’s the type of noise you never notice until it’s gone. Yes, a mini-Itx allows for aftermarket USB or SPDIF cards, but they also require a stronger power supply, with its issues of noise and cost both for the computer and power supply. If cost is no object, you can find lots of threads here pursuing best sound down that path. But I wanted a one computer solution, headless, that could use a high quality ultra low noise power supply that didn’t need to handle 4A or more. So I looked for an i3 or i5 NUC that had a low power requirement, less than 26W, so I could use a 19V 1-1.5A linear power supply. Since I couldn’t find any fanless cases to fit a 5th generation NUC, I chose a refurbished 7th generation i5 NUC. I also needed a NUC spec’d to support two internal drives rather than one. My cost was £200 between the NUC and an Akasa Plato fanless case. Running Euphony Stylus, it never even gets close to 1A except during the boot sequence. Power Supply General consensus seems to be to get the best quality low noise power supply you can afford. It makes a big difference to the sound quality. As mentioned, I wanted a high quality 19V linear power supply that could well handle 1-1.5A and chose one based on the Studer900 PSU phantom power design. That has excellent low noise performance and cost around £150 including the board, case, transformer, and supplemental capacitors. One could buy it complete on Ebay for less but suspected it might not be properly spec’d (espec. transformer) to support this voltage/power combination. SSD & Local storage One friend with loads of experience in the very high end (design, use, measurement) recommended I store the files on an interior drive instead of a network server and that I use an SSD rather than an HDD. This ran counter to most of what I’d read, since SSD’s are thought to be electrically noisier and best to get file storage away from the player. But he said that an SSD requires less current than an HDD and so allows the computer to generate less noise, compensating for any electrical noise from the SSD itself. I found that to bear out in SQ when I changed from HDD to SSD on the storage drive. Of course, I used separate drives for OS and file storage. He also said that wired networked devices will tend to add noise to the system unless a lot of care and expense is put into galvanically isolating them from the player. So, I used a dedicated WiFi 4G router just to enable use of a remote and to have access to the internet, required by some player software. I again found his recommendations to bear out in the resulting SQ. Fanless case & USB2 I moved the NUC into an Akasa Plato fanless chassis, which had a good result to SQ in addition to now being physically silent running. The Plato also allows connections from the USB2 options on the board and I found those to have more accurate transients than the USB3 ports either front or back. CPU Frequency I’ve also heard from multiple sources that setting the CPU frequency is important to the SQ performance and I found this critical in order to get close the disc spinning performance. Fortunately, Euphony’s Stylus software allows easy switching of CPU frequency on the fly for immediate comparison. In my comparisons, changing the frequency could affect: transients, sharpness, midrange, handling of complex passages, dimensionality of images, timing, etc., and I found that the best setting often changed when I changed anything else on the computer and required a lot of comparison testing. For instance when in the original case with a fan, the best CPU Frequency seemed to be 13000 and the timing sounded off when below 10,000. But when in the fanless case, 6000 sounded best in my setup. Status Update Frequency/Processing In Euphony, one can change how often the song position is updated when playing. There was a marked improvement in SQ when it was set to less often and I settled on “never”. This was nearly as significant as the CPU frequency setting. (Euphony: Settings/ Music Service/ Playback Options/ Song Pos Update). Similarly, if using JRiver, the volume control should be Disabled and DSP bypassed. Hyperthreading & Turbo I did not control my experiments optimally but it seemed that it was better with Turbo on and Hyperthreading off. It’s worth popping into UEFI/BIOS and trying them both ways.

I'd already read that. But if 2 songs buffered plus Stylus' foot print = 8GB*, that seems much larger than implied by other conversations here that seem to indicate 4GB is fine for some uses of playing 16/44 music. If 8GB is used when 2 songs are buffered, why is no more used when an entire elbum is buffered? It should be 500MB more RAM used. I have 16GB RAM and it says 7.6GB remains after buffering, which implies that 8GB is used.

I'm confused about the amount of memory used by Stylus. I'm currently running Stylus on a single box system with 16gb RAM. Whether I buffer 3 songs or en entire album, it stills says 7.6GB of memory left. So apparantly I don't understand what its talking about... or how much RAM is used between Stylus and an average 16/44 album's worth of music buffered to memory. BTW: I've nothing stored in the Expert settings about RAM.

Here is Euphony Support's response: "...we advice you to use SqeezeLite with LMS if this works for you. Endpoint is meant to be used only with our Stylus software, and here we give assurance it works in full performance. All other options are only optional, as do not study how LMS is working. LMS is designed to control Squeeze devices which Endpoint is not." I'm not impressed. The Euphony control app clearly shows that, "StylusEP implements two endpoint protocols: Styles protocol... [and] Squeezebox Slimproto-Stylus EP can be controlled by Roon or LMS." [no caveats listed or implied]. There other website language implies that Stylus offers unique SQ (that other software does not). I'm not inclined to pay two subscriptions to Stylus just to get Stylus on the server and the endpoint.

BTW: the probnlem does not happen when the Euphony Endpoint is set to Squeezelite. Before it was an Endpoint, I had it set up as regular Stylus with Buffer before play. That setting is not available under Endpoint but I wonder if it is trying to use that hidden setting and messing things up... and if so, how to test/fix it.

I've decided to try Stylus Endpoint fed by a PC running Logitech Media Server but am having a problem with syncronisation of controls and display Pressing Play, the song timer starts but no music. After many seconds the sound starts coming out at the start of the song. But the display shows I'm 10-80 secibds into the song. The amount of delay depends on the lgnth of the song. If I press pause, the time indicator switches to the correct time for the music that was finally being played. These delays and out of synch display continue on subquent songs in the playlist. Sometimes the player seems to get confused and restarts playing the song even though it's on pause. I'm not really sure where to start in troubleshooting this. I'm using the latest LMS on the server PC. The server, endpoint, and remote app are all on the same router with nothing else on that router. Web Interface: Default Streaming Method: Direct Volume Adjustment: None Database Memory Config: High (8GB hardware RAM)

Yes, I'm aware of those factors. I just find it odd that they alone might cause an ethernet switch to affect the SQ of the music. It does seem that the app goes to sleep when the mobile screen is off for a minute or so. One would think that would be an antidote to this scenario. BTW, the "play and relax" sounds nice until I tried it a few times. Sometimes you just need to be able to pause... and pay any SQ penalty.

After getting an improved LPSU I'm back to testing Euphony through a NUC and a Mutec DDC vs a CD transport, with mixed results in SQ. One thing I don't understand is how the sound would change between being fed by a media bridge router directly vs. putting a switch in between the router and NUC. The files are on an interior HDD, not streamed, and the ethernet connection is only to allow use of the remote and validate the Euphony license. Perhaps because there is regular communication between the player and the remote and Euphony licensing, that somehow causes the traffic to have a different effect on SQ with a switch vs without? Strange. Strange too, the separation of instruments is a bit better with the switch and yet the image depth is lost. So I can hear each instrument more distinctly but it is dimensionally only 2D rather than 3D making it less realistic, no breath of life. The 3D is a bit better without the switch. Perhaps that is a function of the switch and its LPSU not having been used in a while? But after 15 minutes I gave up and removed the switch. In any case, the CD transport still has more realistic dimensionality, tone, flow, and breath of life even if the Euphony-NUC player has a better detail, texture, and separation in the lower mids and upper bass. The CD also handles the background in complex passages better. Perhaps that is down to being constrained by a basic NUC rather than a pimped and fettled PC (working to a budget).

So when the 1135G7 specs say it supports Optane Memory, it means hybrid? "Intel® Optane™ memory is a revolutionary new class of non-volatile memory that sits in between system memory and storage to accelerate system performance and responsiveness. When combined with the Intel® Rapid Storage Technology Driver, it seamlessly manages multiple tiers of storage while presenting one virtual drive to the OS, ensuring that data frequently used resides on the fastest tier of storage. "

If booting to RAM, then in theory an Optane drive would have little further effect I guess. If booting to an Optane drive but buffering files to RAM, then the Optane relates to OS/software related overhead and noise and perhaps cache but not to the difference bewteen buffered & non-buffered files. I didn't realise that Optane as a drive is obsolete. I've been seeing some for sale and thought I'd seen 11th gen NUC's spec'd to be compatible. (my budget lives closer to the NUC world than more configurable ITX, etc.)

What is unclear to me is if I set music to buffer to RAM and get an industrial grade Apacer RAM, does Optane do me any good, or are they alternate approaches to SQ, practically speaking? Separately powering HDD/SDD or RAM is not an option for me.

Also AFAIK, playing a CD does the read error correction real-time whilst playing from a file has had the error correction already done during the rip, if well configured. I use DBpoweramp to rip CD's and not all CD/DVD drives are equal in their ability to read through media errors. Of course opening and send a file and getting it to the DAC without also sending analogue noise down the commons is the bigger deal many of us are pursuing.

Moving to Europe for better HiFi... now that's commitment to the music! LOL Actually it's Tony @Coherent Systems.

A mate has a 4G router dedicated to his HiFi so that it is completely independent of the internet and routers serving the rest of the house. He says the SQ improved significantly just from that. I'll probably try that eventually. I assume you are still employing hardware and software optimisations directly on your server?

With ramroot off, I liked 1.6GHz and Governor=Conservative. But with ramroot, those settings were a bit too sharp and yet the music's sense of space was not as good. With ramroot on, i prefered 1.2GHz and Gov=userspace or perhaps 'schedutil'