Streaming, Optimization and compromises

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'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.

@sns I 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.

@sns That'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.

The NAS complicates things. Perhaps try this:

Ethernet from modem into Buffalo BS-GS2016 switch

Fiber optic from switch into NAS with SFP port

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.

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.