Best Ethernet Cards for Streaming

[jabbr]

Are you really testing 1Ghz cards with a 1Ghz scope ... 🧐

 

Why not just go optical and not worry about these details .... 100Gbe fiberoptic is now at the $1k for switches...

[jabbr]

On 7/28/2022 at 5:45 AM, TomJ said:

This is also what Intel does and states in its test manuael for PHYs.

But perhaps the audiophile community here has developed a rocket science that was previously hidden from Intel.

But I only care about 100base TX. 

 

Ah - clock phase noise Measurements etc. This is all Irrelevant, because what counts is what comes out of the boxes and reaches the recipient. And that is the differential signal and common mode. There is nothing else, if we assume that there is no GND connection via a screen.

 

Oh ok because 100base-Tx is 100Mhz and yes a 1Ghz scope is ok for that ...  that specification is circa 1995 and 1Ghz scopes were considered fast in those days. Does Intel still make 100base-X NICs?

 

So no nothing is hidden from Intel and in this century Intel is concerted with silicon photonics eg

https://www.intel.com/content/www/us/en/architecture-and-technology/silicon-photonics/silicon-photonics-overview.html

https://newsroom.intel.com/news/intel-demonstrates-industry-first-co-packaged-optics-ethernet-switch/

 

As you can see Intel's current lineup includes 100-400 Gbe

 

This isn't "audiphile community rocket science" rather mainstream networking rocket science.

 

Now ... all this said your measurements seem to support the idea that common mode noise transmission is the primary factor affecting Ethernet "SQ" which is what my much more limited testing first pushed me to use fiberoptic ethernet. I am surprised the common mode noise values are as high as they are (100-200 mV) despite the Ethernet PHYs and I can imagine they might be even higher in some circumstances. Those values are indeed measurable!!!

[jabbr]

4 minutes ago, TomJ said:

 

Dont want to go into this discussion, because i have done this too much, but 100base TX works with Multi Level Transmission MLT-3 so the bandwidth isnt 100MHz . . . . can google about this and we should not spam this thread with this well known stuff.

 

Yeah, typo, I meant 100MbE or 100 megabits per second Ethernet, not 100Mhz. Regardless a 1Ghz scope is not adequate to make jitter measurements on 1 GbE which is the common speed used on home ethernet. Intel doesn't currently use 1 Ghz scopes to test Ethernet regardless of what they wrote in the previous century.

[jabbr]

23 minutes ago, TomJ said:

You are wrong - read the linked Intel paper in my thread and you will see that they used the same oscilloscope and the same probes as I for 1000base TX. Its not the newest paper, but they have done this.

I'm sure you know that the testing standard was entirely revamped between 1Gbe and 10 GbE (10Gbase-X) in which the stressed receiver jitter testing was introduced. Sure you can do basic compliance testing for 1 GbE but that doesn't mean that you can do accurate jitter measurements ... its not just the speed of the scope

[jabbr]

23 hours ago, TomJ said:

And common mode noise will not much differ between the bandwidths - but you will know all this, because you are for sure an expert of ethernet jitter and common noise measurments . . . 

 

Your measurements of common mode noise seem rather high at 180mV and even at 90mV for the EtherREGEN port. The noise/jitter allocations for each step in the chain go down as network bandwidth goes up e.g. at 100Gbe its even stricter. I doubt those levels of noise would pass compliance.

[jabbr]

26 minutes ago, TomJ said:

As I wrote in the posts of my measurement Thread: The measurements are only valid for comparisons. Not absolut regarding the values.

I hear you. EMC/RF shielding to do proper common mode noise testing can be difficult -- need faraday cage/lab. I have said many many times that my opinion about the reason for different "SQ" of different copper Ethernet devices is likely due to common mode noise and not even "jitter". The point about 10Gbe isn't that this speed is necessary, rather that starting with this standard. compliance testing ensures that jitter and noise are not passed across network hops and do not accumulate. The point about fiber isn't that the bits going to the DAC are better because they aren't, rather that fiber doesn't transmit common mode noise.

[jabbr]

On 8/4/2022 at 3:39 AM, TomJ said:

As described above, although I do not have a shielded measurement room, the measurements are comparable.

But your thesis regarding CM-noise agrees with my findings.
If I understand you correctly, you say that at 10Gbe the requirements for compliance with CM-noise are higher and thus it is ensured that network devices produce less of it - right?

 

With fiber, I have the following concerns regarding the argumentation: 
If a switch is well designed in terms of Bob Smith termination and additionally grounded, very little of the noise upstream of the switch will reach the network downstream.

I have not yet tested a fiber-copper converter, but there are statements from developers that the fiber components are strong noise producers.

 

I am saying that fiberoptic networks do not transmit common mode noise through the glass ethernet cable. I've done no comparison testing of 10 Gbe copper, nor recommend it due to the higher power needs.

 

10Gbe compliance testing ensures that upstream noise, is not retransmitted downstream, whether that be common, differential noise nor phase noise. More precisely this is defined in the "stressed receiver" testing. 10 Gbe and newer uses end to end testing.

[jabbr]

38 minutes ago, TomJ said:

Here is what I said:
There haven't been measurements of fiberoptic converters so fare. So I can't say with 100% certainty that a network galvanically isolated by fiberoptics will have less noise at the endpoint - that's all I wanted to say.

The biggest problem is the switch itself as a noise producer, less the upstream in my opinion.

Will look at the 10GBe stuff. But if BS is implemented proper for each port, then in my measurements there is nothing I can see critical from upstream noise in downstream.

 

You were specifically discussing common mode noise transmission over an ethernet cable from different ports on a switch.

 

common mode noise simply is not transmitted over a fiberoptic cable. The fiber could transmit intensity noise and jitter. Both intensity noise and jitter would be seen on things like an eye pattern. The 10GbE+ specs do specify eye patterns.

 

Common mode noise at a fiberoptic endpoint would arise from the endpoint itself, not a port on a switch (the switch could emit noise over the power lines that could couple to the endpoint power supply -- i do use a robust isolation transformer in my audio area to prevent coupling with other electrical devices such as refrigerators etc)

 

The newer ethernet specifications e.g 10Gb and newer require **noise rejection** such that noise may not accumulate across hops -- otherwise the network simply wouldnt work and the internet infrastructure is 100Gb and up these days...

[jabbr]

From my POV I'm more concerned with the amount of noise at the endpoint than the switch (because I use 10+Gbe fiber switches) so compare the noise in the Intel x520 vs Mellanox vs OpticalRendu vs ClearFog ***

 

Its the endpoint that is electrically connected to the DAC ... typically via USB

 

*** I have no affiliation with ClearFog, just that circa 2017 it was the available low powered endpoint with an SFP port

[jabbr]

5 hours ago, TomJ said:

At the end of the day it is important what noise will be carried through USB to the DAC. Is there any information if clearfog has put effort here?

 

That would be a fruitful area to measure. You could look at different endpoints. You could look at different network inputs e.g. copper vs fiber and see the effect on the noise at the USB output.

[jabbr]

5 hours ago, Jud said:

In that light, @TomJ isn’t wrong to be interested in the noise a fiber to copper converter inside a switch may be putting out. It wouldn’t be a bad thing to do some measuring, and also not a bad thing to put out information about various best practices for keeping electrical noise generated by network components from affecting the eventual output.

As I said just above ... the common mode noise at the USB output is what I'd be interested in.

[jabbr]

22 hours ago, TomJ said:

At the end of the day it is important what noise will be carried through USB to the DAC. Is there any information if clearfog has put effort here?

 

You know, if you can do your common mode noise measurements on the USB output signal, this measurement could be a reference for a whole host of factors, including endpoint hardware, power supplies, network hardware etc etc etc -- it would be really great, and at this point we could discuss the intricacies of common mode noise testing. You could even measure the USB CMN at the port vs at the end of various USB cables.

 

People could also repeat it. I do strongly suspect that "SQ" would correlate.

[jabbr]

1 minute ago, Superdad said:

 

@JohnSwensonwill be including a bunch of information on this very subject as part of something we are preparing to help guide @TomJ's investigations.  

The place the CM noise matters most is at the differential between the DAC clock's input pin and the ground plane. And the frequencies examined are important also (so far Tom/Eric has been looking much too broadband; 1MHz and far below--especially way down at AC line frequency and multiples--matter more).

That's all I'll comment on at the moment. Will leave the rest to JS.

 

No doubt. Some actual information is better than none!

 

Ultimately the inputs th the DAC, no just the clock, matter. There is perhaps too much obsession with the clock itself. In any case that measurement is a hella lot more difficult to make, than simply the USB outputs and we have been waiting for measurements for X years now...