Networked DACs and network speed?

[Miska]

16 hours ago, miguelito said:

I had read that network speed should be fast enough but not faster. The idea is the higher the speed, the more electrical noise generated. In this vein, I see both the Melco audiophile router as well as the EtherRegen have connections to audio endpoints at 100Mbps rather than 1000Mbps.

 

So first question: Has it been established that connections at 100Mbps are better for audio?

 

No, faster the network, the better. Faster networks give you shorter data bursts (and thus less radiated noise energy). So 1 Gbps at minimum and nowadays things are moving towards 10 Gbps networking.

 

And if you want to use something like RAVENNA, 100 Mbps won't cut it.

 

For networking, I rely solely on HPE and Cisco gear. Two central patchbay switches, one Cisco and one HPE. Then each room has it's own room specific switch which are all HPE. Wireless access points are 802.11ac and powered using PoE power from the central patchbay switch, so only single wire going to each wifi AP.

 

[Miska]

8 hours ago, plissken said:

So if you have a playback device that can cue up 700MB of information, basically an hours worth, and the NIC go to sleep for the duration and even lending any credibility to lower speed = better, wouldn't the highest possible speed interface actually be the best?

 

Networked DACs don't hold an hour worth of music. And let's say you have a Merging Hapi with 8 channels of input at DSD256 and 8 channels of output at DSD256, or 16 channels of output at DSD256. You have have 172 Mbps worth of constant traffic.

 

Likewise, if you play to exaSound 8-channel DAC at 384/32 you have 93 Mbps worth of constant traffic.

 

Neither one is going to work at 100 Mbps.

 

Higher the network speed, longer the sleep periods are between the data transmissions.

 

[Miska]

12 hours ago, k-man said:

Darko's article "Gordon Rankin on why USB audio quality varies" - something like 'the faster the interface the more chance that there will be an error'.

 

You can always check error counters, if you are worried...

 

For example in this office room switch things have been running smoothly:

 

12 hours ago, k-man said:

The B side 100Mbps port which allows the ADIM (galvanic moat) is the only 'affordable' piece available. The 1000Mbps either doesn't exist yet or too cost-prohibitive at this time.

 

Just use 1 Gbps or 10 Gbps optical Ethernet and you don't need to worry about galvanic isolation. Not cost prohibitive and widely available in regular networking gear.

 

P.S. What is "B side" port for Ethernet? Such only conceptually exists in USB.

 

[Miska]

1 hour ago, DavidL said:

For what it's worth when considering the purchase of the original Sonore Rendu (ethernet > SPDIF) I asked Jesus Rodriguez whether I should use GB ethernet or 100Mb ethernet output from optical isolation to the Rendu. He said the latter as this would reduce the potential for RF interference.

 

I would like to hear what is technical basis for such. Transmission times are longer with 100M, signal levels are the same...

 

[Miska]

3 hours ago, Dutch said:

higher bandwidth (spectrum wise) in use for gigabit
4 pairs vs 2 in use in the network cable

Different mode of operation of the PHY

 

Gigabit uses primarily more complex signal, a bit like if you compare 2G/3G/4G mobile networks; frequency band usage is largely the same, but same frequency band can carry more information due to larger constellations and advanced signaling methods.

 

While higher frequencies have less penetration than lower ones.

 

3 hours ago, Dutch said:

Higher load on all circuits following the PHY including but not limited to CPU and memory could cause more processing noise

 

I would say unlikely, because the transfer/idle time ratio is much less favorable at lower speeds. While at higher speeds there is more idle time where the hardware can power down. This is especially notable for CPU and memory. This specifically favors use of higher speeds where you quickly do something and then sleep longer. Rather than staying awake for long periods of time and running inefficient idle loops more and maybe not having time to enter sleep states at all.

 

In addition, many of the power saving features are typically either only available at Gbps standards, or more efficient (802.3az).

 

[Miska]

https://en.wikipedia.org/wiki/Energy-Efficient_Ethernet

[Miska]

52 minutes ago, Dutch said:

one last thing; I know the encoding technique differs and the clock rate may be the same at 125MHz but gigabit ethernet uses a bigger bandwidth compared to fast ethernet. See this for example: https://www.flukenetworks.com/knowledge-base/applicationstandards-articles-copper/mhz-vs-mbits-and-encoding

 

IMO, that is not a problem. Energy wise 10x or 100x longer transmitter and chip activity is more of a problem.

 

Unfortunately Ethernet doesn't use spread spectrum clocking and randomized frequency hopping like military radios for example do.

 

[Miska]

In addition, of course with a good switch and Ethernet interfaces, EEE cable length detection is used and only amount of power needed for the particular cable length is used. For example Cisco switch supports this only at Gbps speeds.

 

It is kind of pointless to blast at power specified for 100m cable length if the cable is just 2m long...