jabbr

For folks wanting low cost yet high performance and modern network switches that ... basically use a modern spec that mandates eye pattern testing for compliance and rejection of upstream jitter: https://www.servethehome.com/mega-2-5gbe-switch-guide-update-with-21-new-models-added/ The added advantage of the newer SFP28 modules is that they have multirate support when used in a switch that supports multirate

I just want to clarify this: The MC200CM does not use SFP modules at all, it directly accepts fiber terminated in SC connectors 99.9% of the discussion here uses fiber terminated in LC connectors which attach to SFP modules.

Let me respond to this: Sure! It is possible that upsteam jitter is blocked by a 10Gbe segment and doesn't have the same electrical impact on the endpoint. This isn't just clock jitter rather the "eye pattern" which doesn't pass upstream noise by specification. There's been a ton of speculation (including by myself) about why this might be the case, we just don't have electrical measurements. 10Gbe might be the "sweet spot" because although the "jitter" goes down with even higher network speeds, there is also packet timing. Higher network speeds mean that packets take up less of the total bandwidth but I've also seen retries go up substantially and tuning the network can be a black art form. My last segment is 10g so ... meh life is short and my hearing isn't able to detect a difference. But *yes* is people generally hear an improvement at 10G then that does suggest netwok jitter may matter though I can't give a very firm explanation. I *have* considered bursting a song at 100G during say the pause between songs so that the network isn't actually active during playback on the DAC but that would be a lot of work for a theory.

There are all sorts of network debugging software in linux that will give you statistics on actual transmission rates, number of packlets that need to be transmitted etc. netstat is popular and I use I can tell you that if you are seeing problem at 10Gbe that those will be much much worse at 100Gbe even though you aren't needing the full data rate. I have no idea if these retries commonly affect audio, but who knows. Network tuning in linux can be an art form!

This is exactly the type of feedback this thread is intended for! Yep a switch is only guaranteed to work with SFP modules on its compatibility list.

Use corning 'ClearCurve" bend insensitive fiber

Posting these articles intermittently as the future of optical networking. This has **nothing** to do with home audio. Intel has previously demonstrated but looks like Broadcom has taken the lead in implementation. The metal DAC cables (direct QSFP to QSFP) and QSFP ports may go away to be replaced by optical to optical, presumably single mode fiber direct from chip to chip. I'd assume the fully optical chip could plug into a QSFP port on a remote device... Broadcom Now Sampling 51.2T Co-Packaged Optics Switch (servethehome.com)

In the context of network devices (this subforum) a reclocker is simply a device, typically a switch but also an extender or FMC for which the output eye pattern is equal or tighter than the input eye pattern.

Yes FMCs are simply very minimal switches with one copper RJ-45 and one SFP port. Using a switch with more ports is typically more cost effective.

Solarflare only has drivers for server versions of Windows. They were not intended to be used on desktop machines. They were successfully marketed at the trading community for which trade execution speed is critical hence their low latency. So you couldn't find a driver for your Windows OS. X2 Series Ethernet Adapters (xilinx.com) It is true that lots of fiberoptic equipment isn't intended to be used in the client/desktop environment and there are consequently more incompatibilities than with desktop products which are more intended to be plug and play. This is the reason for this thread: to document what works with what. I have successfully used Solarflare NICs with Linux. I have found that a number of the server NICs of that generation ran hot and need good cooling. One of the advantages of the Intel fiberoptic NICs is that they list wattage and you can typically figure out how hot a card will get by the number of watts it uses.

1) Solarflare NICs are terrific, arguably the lowest latency 10G NICs as well as supporting the PTP if this can help. Now part of Xilinx -> AMD 2) passive DAC cables aren't fiberoptic, rather copper

ZFS has always sounded best to me because I just don't need to worry about it.

Hey @Jud I see you have a fiberoptic USB cable, what do you think?

At this point all discussion of such noise has been theoretical and despite promises that "any day now" we are going to see data, after 10 years there hasn't been any -- for audio systems. Correct it can't be additive for high speed modern computers and networks. For 10G+ networks there has been masses of data and there are devices which measure this because its a big industry. Differential noise, whether ground plane, supply plane, or phase, isn't allowed to propagate through a 10G+ switch ie it isn't allowed to be additive and there need to be circuits which are designed explicitly so the noise isn't additive. At 100G, for example, even a tiny bit of noise can throw the system completely off. Its not just peer reviewed research, its by specification and design and compliance testing.

Yeah **honestly** I couldn't predict how much input ethernet "jitter" effects SQ on the Lumin. It might not at all because buffers etc. There's been alot of hand wrangling to suggest that it might but in the decade that this has been debated and despite promises of tests etc I haven't seen an actual shred of evidence to say that the Lumin, to give a specific example, SQ is at all affected by input ethernet jitter. Now I've chosen to remove the possibility that it might be important by using 10g switches with a 1g output but really its theoretical and subjective reports which don't use modern highspeed fiberoptic ethernet (eg 10G and faster) It would be interesting to get something definitive from Lumin if they have any data...