So, find out Covid hit some family members, Christmas called off so here at home with time on hands.
First off, this thread not for objectivists for whom general service computer good enough, heard your objections prior, no need to repeat.
Recently I've come to realization my extreme optimized mac mini will become obsolete sometime in near future. So once again investigating future server replacement.
First off, any quality server solution requires good power supply, whether internal or external, pure lps or combo of lps and switching such as Antipodes offers.
Now we come to output, I'm going to concentrate on the two most common outputs available, usb and ethernet. The first is usb, an inherently compromised digital audio interface. So we have a number of ways usb implemented in servers, usb directly off motherboard and various schemes for usb optimization. Directly off motherboard is serious compromise, noise inherent to motherboard, add to that running added processes to motherboard, therefore, need to act as streamer in addition to server, which adds even more noise. Also, using motherboard power supply and motherboard clocking, all are serious compromises. Now, we can optimize usb output either internal or external, dedicated power supplies and clocking can improve usb out. In spite of all this, still have inherent liabilities of usb.
And now we come to ethernet out, a superior solution in my experience. I've bridged thunderbolt as ethernet out on my mac mini, far superior sound quality to the motherboard usb out. While I can't directly speak to superiority of ethernet to usb outputs on all servers, for sake of discussion I'll presume this to be true. My reasoning for this is with ethernet out one has option for FMC (fiber) and outboard streamer. Fiber eliminates all noise prior to FMC and outboard streaming relieves motherboard of those duties and isolates many choices of optimized streamer and dedicated power supplies for those steamers. These streamers will also optimize usb with dedicated clock. Your could also have many options of streamer/dac with ethernet out.
The way I see it, even ethernet out is somewhat compromised in that in order to take advantage of fiber's advantages one must use FMC. I'm now beginning to see fiber output on servers, Sonictransporter I9 is one I know of. Advantage of this is eliminating clutter in form of one FMC and attendant lps, although in my case requires FMC adapter.
At this point, I'd suggest ultimate streaming solution would be fiber out server to fiber capable streaming dac, lowest noise and reduction of clutter. Until this happens I see all present solutions as compromised to some degree. As far as my top choices for off the shelf solutions at this very moment. Antipodes, Innuos both have ethernet out options, both compromised usb outs except on highest level models. Mojo Audio Deja Vu has both optimized usb and ethernet on board.
What do others have to say in regard to my criticisms of usb outputs vs ethernet outs? Have any of you directly compared optimized usb to optimized ethernet outs on same server in your own setup? My take is usb meaningless, and if ethernet out is preferred setup, most server moves will be lateral for me. Doubtless better usb out solutions available vs my lousy mac mini usb outs, again, this not part of my question. Also, I'm not looking at streaming dacs in this equation. I'm happy with present dac with optimized usb in at this point. My discussion is only meant to compare server usb vs ethernet out interfacing with usb dacs.
I'll be building my own music server in the next month. Having enough processing power for upsampling to DSD, and implementing fiber optic are priorities. All components will be in a fanless case to minimize audible and electrical noise.
The server will utilize a Solarflare fiber NIC, and a JCAT Femto ethernet card. I have a streaming DAC with RJ-45 input so will be running cat6 UTP straight from server to DAC. Fiber optic into server from either Uptone EtherRegen or basic FMC. Ethernet cards like the JCAT have upgraded clocks as well as LP4 Molex connectors to be powered by separate 5v power supply. No need for power from motherboard. JCAT, Pink Faun, Taiko all have USB cards with same features if USB output is needed.
RAM will be industrial grade ECC memory. Many disussions over the efficacy of ECC vs non-ECC. I believe most of the reported improvements come from the use of industrial grade memory, which has a wider operating temperature range and higher quality components. ECC less of a factor.
OS will be Linux based. Still deciding between Audiolinux, Euphony, or basic Ubuntu. Audiolinux and Euphony can boot from memory, so any noise from SSD is eliminated. Both also come loaded with Roon and HQPlayer embedded.
Power supply also still undecided. Fanless SMPS with DC-ATX converter on the cheap side, HDPlex ATX LPS more expensive, Taiko DC-ATX converter with unregulated LPS more expensive yet.
For me, building my own server is the way to go. Look inside most of these high-end bespoke servers and you will see motherboards from Asus or Supermicro, and processors from Intel or AMD. Their proprietary OS's are all Linux based. These servers are merely computers at the end of the day, most built with off the shelf components. DIY saves cost and allows for total customisation. You have freedom to switch components and software at will.
Of course this option not ideal for people with no ability or interest in DIY. For them a turn-key solution from any of the top brands is ideal. Nothing wrong with that. It all accomplishes the same thing.
@sonic79I'm leaning diy route as well, doubt any of the moderate priced off the shelf better what I have. What motherboard would you be using? I too have heard the industrial grade memory and ssd preferable. I wasn't aware Roon embedded in that software, good to know!
Scratch build would be way I'd go too, none of the easier diy builds impress me thus far. My mini tower computer is scratch built, far more capability for the money than any off the shelf variety.
@snsI have most of the parts list compiled. Like I said I'm still determining the OS and power supply. The hardware aspect is simple, basically a standard PC build. The software is more advanced. Audiolinux is based on Arch Linux which requires a bit of manual setup. Euphony is more turn-key, but also more expensive.
Processor will be an AMD Ryzen 5 5600X. A fanless case that cools the CPU passively limits the choice of processor. No fans to move air so we're relying on a heatsink. The 5600X is rated at 65W TDP, while comparable Intel processors are 105-125W TDP. AMD has a big advantage over Intel in efficiency. Running the more demanding noise shaping filters in HQPlayer requires high base clock frequency and large L3 cache. The 5600X does well in both areas. If you don't plan to use upsampling/DSP, an even less powerful processor can be used.
Motherboard will probably be an MSI B550M Mortar. I am going with micro-ATX because the case I intend to use is too small for full ATX. No worries here, as micro-ATX offers all the features I need. This MSI board has great VRM performance. If you want the most advanced VRM look into the Gigabyte B550 Aorus Master, or X570S Aorus Master. They have 16-phase design. Both are full ATX, so the appropriate case is needed. All these boards are designed to accept AMD processors, so if you intend to use an Intel processor you need to research boards for that application.
One of the most expensive and highly regarded servers is the Taiko SGM Extreme. It uses an Asus WS C621E Sage motherboard with dual Intel Xeon Silver processors, industrial grade RAM, custom internal LPS, all passively cooled. There is a huge thread over at Audiophilestyle dedicated to building a DIY version of it. Cost can get extremely high depending on the hardware you use.
@sonic79Thanks for the info. Not sure if I'll end up using HQPlayer , can't experiment with it with present mini os, one of the issues forcing my hand on new server.
@gdnrbobThanks, I'm aware of audiophilestyle, been lurking over there, was was source for most of what I learned in mac mini build. Have to check out WBF more in depth.
@snsThat's one of the main reasons that prompted me to build a server. My current computer is too old to update to newest version of Mac OSX, which is required for HQPlayer. Roon has upsampling, and I have compared native PCM to upsampling to DSD128 (my DAC is limited to DSD128 over PCM). I prefer upsampling to DSD, hence my requirement for adequate processing power. PCM is more forward sounding with more thickness in the midrange, but has a level of fuzziness. DSD pushes the soundstage rearward and removes this fuzz. Overall sounds cleaner, more air, with more perceived resolution. But this is a discussion for another topic.
HQPlayer has even more powerful upsampling than Roon. The player is supposedly better sounding than Roon, as well. I'm sure you are aware of the benefits of upsampling to DSD external of DAC. Delta-sigma DAC's like your Musetec automatically convert PCM to DSD internally. Powerful software like HQPlayer does the conversion external of DAC, and does so more effectively. This is a process removed from the DAC, instead done by the server which is more suited for the role.
Well, I certainly don't like Roon dsp, reason I wanted to try HQplayer. Also want to give Audirvana Studio a try, slightly preferred sound quality of older version Audirvana vs Roon back in that time. Audirvana requires newer os as well.
My mini can be upgraded to latest mac os, my issue is since so many processes disabled worried about os upgrade. I may have to go back and undo all the optimizations, including dos command optimizations to have glitch free os upgrade.
You will want to look at the company small green computer. They have already addressed these issues. In their process they have concluded that it does not matter how much noise and rf are in the output as once you feed it into fiber as part of the output it is eliminated.
@sns With your building comfort/experience, DIY will give you the most performance for the money. In my case, saving time and effort led me to SGC i5 and optical rendu to take advantage of the fibre benefits you mentioned. It works and sounds great, but compared to DIY builds using premium parts, you'll save a bit doing it yourself.
+1 @gdnrbobthe crowd on ..style will provide far more experience on things like comparisons of motherboards. DIYaudio might be another good source.
I've recently read that fibre cable can go very long runs without picking up noise or degrading quality at all. If FMC in an audiograde switch with FMC and running fibre cable through the house is an option, that might be worth exploring towards your assertion of "ultimate will be..." Cheers,
@sgreg1On one hand it makes sense, noise and rf prior to fiber wouldn't matter. On the other, makes sense noise anywhere along the network would make a difference. Lets say one had fiber internet provider, so full fiber network, noise eliminated between every link in chain, and then pay attention to noise production in the components themselves, logic states this would be optimum streaming setup. It makes sense noise produced within each component affects the processes taking place within that component.
Also, if noise produced within components, server in this case doesn't matter, why does small green computer even produce a audiophile server, wouldn't a general service computer work just as well?
I've used a variety of servers over the years, my experience is that servers don't all sound the same. The extreme variety we observe in streaming setups is empirical evidence there isn't an objective optimal setup for streaming. The only objective goal I have is to minimize noise at every point in entirety of streaming network, we do have the means to accomplish this. I hear noise diminishment in my chain as higher resolution. My streams even with present non optimized setup sound essentially same as cd rips. And my rips were superior to cd's played over transport some years ago, there's only been improvement in my streaming since that point. This all due to maximum attention paid to noise suppression on my total network.
@sbankIn regard fiber cable, that was my understanding as well. I have considered everything post modem, the only things we have control over in thinking about optimization of streaming. I did try an audiophile switch between router and server/NAS, didn't work well for me, I theorized excessive clocking in my case. It may also have something to do with my attempts to optimize my router with over spec LPS powering it.
DIY is highly likely to be my choice, I just don't want to spend a lot of time with researching and choosing parts list. If I can get total parts list together, I'll have fun with the build part of equation. I did do all the research and parts choosing with my diy general service computer, rather exhausting part of process for me, I like the actual build much more.
I recently recall seeing a Taiko server copy build, quite amazing, not sure I'm reaching that high.
For most of us the best we can do is to convert to fiber as early in the chain as possible. In my case I have coax running into house from the road, and so the earliest opportunity to convert to fiber is after the modem. Better yet would be having 10GbE service so you have a fiber modem, but availability and cost of fiber service is detrimental. I researched coax modems with SFP ports without success. It makes sense why such a thing doesn't exist, as the two mediums are contradictory. This means next best option is either FMC or switch right after modem.
Fiber optic cable is so inexpensive that converting to fiber early in the chain actually saves money if running long lengths of cable is needed. A 20m fiber cable is under $20, while cat6 UTP could be over $100. Then the noise rejection property of fiber is enjoyed. A long run of UTP snaking through the house has many opportunities to collect RFI.
One last bit of advice is to avoid going through a router if possible. This is one of the most extreme sources of RFI in a household. I had to plan a chain that completely bypasses the router.
I've thought much about entire stream chain. I don't have fiber internet provider, so I run modem relatively close to system, ie longer coax vs. ethernet cable. So short AQ Vodka from modem to router, router w/lps short run aq vodka to router and NAS. I tried the audiophile switch after router, no good for me. But yes, I agree FMC between router and server/NAS would be definite upgrade. Issue is four FMC, four lps for FMC unless I was to run cheap switching ps. I could go with two FMC for server only, keep NAS direct from router, so streams would be optimized, NAS not.. There is no doubt router to server and NAS weakest link in my setup.
Now, I do wish I could do without router and wifi altogether, I could make this work with much inconvenience. Same audiophile switch I don't like could work in this case. In fact, I will run temp experiment with this setup just to see router's influence on sound.
I do take emi/rfi seriously, have Trifield meter, have provided rfi shielding where necessary. You wouldn't believe how much rfi comes off transformers, they are major contributor to noise, wifi needs no introduction, terrible thing. Anyway, I've had to make compromises, just as we all do in this realm. This emi/rfi thing is why I like totally discreet streaming setup, no streamer dac, no server/streamer, no displays.
Second fiber optic cable from switch into server with fiber NIC
Either ethernet from switch into SOtM, or ethernet from server into SOtM
-OR-
Fiber from server into OpticalRendu or SignatureRendu
Retain router purely for controlling the server headlessly
This would require buying a BS-GS2016 switch and a new NAS with SFP port. The new server you build should have a dual port fiber NIC. You completely bypass the router (WiFi only needed to connect your tablet used to control the server headlessly). The Buffalo BS-GS2016 switch is a popular choice for a high-end switch. RenoLabs, Melco, AfterDark, and some other brands all use the BS-GS2016 PCB. You can easily modify it to be powered by a LPS. You would then have a top notch switch for not much cost. They are more difficult to find these days since audiophiles discovered it and bought them up.
Buy a new NAS with SFP port to allow for straight fiber connection. No need for a FMC. Alternative option is to transfer your local files to a SSD inside the new server, eliminating the NAS. The SSD can be powered by a dedicated 5v source.
Consider replacing the SOtM with an OpticalRendu or SignatureRendu, both of which have SFP ports. This eliminates another FMC if you want a fiber connection to your streamer/Roon endpoint.
I would try to simplify things as much as possible. Having FMC's and cables and power supplies all over the place would drive me insane.
Scenarios you've laid out are good, exactly what I'm looking for with simplified nearly all fiber setup.
Yes, I have complex setup, something like ten steps just to get system up and running. Sounds good as is, assume potential for so much more. I'm not sure that streaming has any more inherent limitations than what's on a cd, perhaps it has more in that we can virtually eliminate jitter vs. cd playback.
Streaming definitely more complex undertaking than CD transport. That's why I place a premium on simplifying the network chain as much as possible. Fewer links means fewer weak spots. It can get overwhelming otherwise.
If you moved your local files to a SSD in the server, this would free up the second SFP port on the switch. No more NAS so one less box, one less LPS, less cables, less variables. Pretty streamlined chain.
Modem into switch
Fiber from switch into server
Fiber from switch into OpticalRendu or SignatureRendu
USB into DAC
Of course we could then have debate over NAS vs internal SSD. On one hand NAS offers complete isolation from server. On other hand internal SSD is simpler, and can be powered by dedicated 5v supply to isolate somewhat. The question then is the NAS going to be any less noisy than the SSD? The NAS itself is a SSD. If the answer is no, then we're adding unnecessary complexity with the NAS in the chain.
Almost all of the high-end servers available have internal SSD's for the OS and local files. The designers of these servers must not be too concerned with noise from the SSD, unless they are offering a SSD purely for convenience. Hard to believe that with some of these servers like the Taiko Extreme. No stone left unturned there.
+1 on the Small Green Computer references from @sgreg1and @sonic79
There are many moving parts across this subject and it's evolving at a tremendous pace, literally almost daily/weekly
One would like to think that networking standards and principles across data, voice, video and now hi-res audio would be consistent - but all the audio grade network equipment and cables on the market suggest that's not the case
Finding the sweet spot these days can be a who's who in the zoo mixed with a multitude of snake oil conspiracy theorists
@sns, sounds like you would benefit from an email exchange or phone conversation with Andrew Gillis
He's the founder of Small Green Computer, knows most of answers to your questions off the tip of his tongue and equally concerned about getting you the right information to make good decisions as much as he is in making a sale
I'm extremely pleased with the product performance and service from SGC
You must have a verified phone number and physical address in order to post in the Audiogon Forums. Please return to Audiogon.com and complete this step. If you have any questions please contact Support.
