A better way to stream

A streamer as endpoint would be applicable in case of Roon. Other software like Euphony Stylus can also benefit as well. Depending on streamer, many other functions may be available which also may be beneficial.

The other advantage as I stated previously is ability to bypass the contentious network switch. Another possible advantage is superior clocking that may be incorporated in renderer/endpoint/streamer (goes by many names and functions).

The greatest possible disadvantage of streamer only setups is likely usb out, usb can be much improved by the rendering provided by many of these outboard devices. I observe atx servers provide much better usb rendering within, usb boards like Pink Faun, Jcat. How much rendering are these off the shelf servers offering?

Perhaps Innuos amd some others playing marketing games stating this finding. Yes, many are finding FMC can be both high value proposition and/or most effective means to lower noise floor on network. Optical provides complete isolation from noise caused by upstream components. EMI/RFI both generated within and bombards network equipment and following on grounds is completely eliminated, filters only filter.

In the end, only listening will determine value of optical for any particular setup. Two generic FMC and optical cable can be tried for minimal cost. Optical conversion is easily most universally positively reviewed network component I'm aware of.

If Node you're all in one unit, not much to do, you can even use wifi.

Something like NUC's with separate dac are next step up from Bluesound. At this point network optimizing can pay off. Basic setup will usually be server/streamer all in one package with usb out to dac, usb renderers are first optimization. The issue with most usb outs is they come directly out of motherboards, not ideal, reason why people find usb renderers effective.

Next people generally move to network optimization, which means fancy switches, filters. I'm just proposing another path for optimization, simplifying while employing optical conversion.

My proposal for  optical with shortest signal path would be optimized optical out on server to optical usb renderer/end point/streamer (whatever you want to call it) to dac. This is optimal isolation, read lowest noise floor, should be highest resolution streaming solution.

Since optical out servers are practically non-existent one has to do the fiber media converter dance,

Some people want to stream for convenience, others seeking highest possible sound quality. Many good choices for convenience, have to get into weeds to achieve highest sq.

My take is most off the shelf servers not for me, too little bang for buck, no two ethernet out (with exception of Antipodes, SMG), don't render usb internally and only the very highest end ones will do HQPlayer high level processing. One can do much better with diy atx build, possibly NUC build. 

Highest quality streaming dac can also be another top choice, forget the usb altogether, all you need is optical for the isolation. On some dacs I2S may also be best input, again don't have to worry about the usb stuff, optical is all you need add. Unfortunately, most dacs optimized for usb, rendering and optical both needed for best streaming quality..

@fuzztone Neglected to mention Small Green Computer Sonictransporter I9. Checks nearly all the boxes, Sonore Opticalmodule optical out, fast Intel I9 processor, should be able to run high level HQPlayer dsp if desired. Also has on board storage in various capacities.

https://twitteringmachines.com/review-sonore-signature-rendu-se-systemoptique/

This example of what opticalrendu product can do for system.

@fuzztone Playing a hard drive on network, NAS. If hard drive in server you have to go through network switch.

If not using fiber out from server, things get more complex because the conversion required. However, the ultimate optical setup can be pretty damn simple, server with optical out>optical in usb renderer/streamer/end point>dac. Three items, you could cut it down to two if there was such a thing as streaming dac with optical input.

One of the major points I"m making here is the liability of vast majority of usb outs on servers. Many have discovered improvement by using various usb renderers, filters, etc., this complicates what should be simple. An atx motherboard build with Pink Faun or Jcat usb board takes care of the usb problem, but at a cost, $500-$2,000, then add need for dedicated lps for these boards.

So, there is less than optimal usb out and not enough processing power for those who wish to use full potential of HQPlayer dsp. Two inherent flaws of too many off the shelf servers.

Also, there is hierarchy of optical products as well, at low end the generic FMC like TPLInk powered by smps, next up is products like Sonore Opticalrendu, these have Femto clocks and powered by LPS, at the top are Sonore Signaturerendu SE.

My recent upgrade from TPLInk FMC, both powered by lps to Opticalrendu powered by Uptone JS2 was not a small upgrade. Heard within seconds, another layer of noise removed, this easily one of the most meaningful streaming upgrades I've made in four years of streaming.

@sdl4 Perhaps Innuos is referring to the cheap FMC when speaking of adding noise. Adding lps to these certainly lowers noise floor. Sonore observed some modifying internals on these generic devices, thus, was born inspiration for the Opticalrendu line.