Talk:Terabit Ethernet

revert of "used to describe speeds of Ethernet at or above 1000 Gbit/s"
The reasons for reverting Madbrenner's good faith edit from "used to describe speeds of Ethernet above 100 Gbit/s" to "used to describe speeds of Ethernet at or above 1000 Gbit/s": --Zac67 (talk) 16:53, 13 July 2017 (UTC)
 * 1) There is no 1000 Gbit/s Ethernet today or in the near, foreseeable future. 200 and 400 Gbit/s are in development and expected by the end of 2017 while true terabit will take at least another 5 to 10 years (my personal estimate).
 * 2) The title orginated from early 2010s efforts to develop a terabit speed Ethernet PHY. However, as of 2017 this has only led to the development of said 400 Gbit/s due to technical feasibility, expected pricing and expected market acceptance. This is already sourced in the article but not obvious – room for improvement.
 * 3) It may be necessary to discuss whether the article has to be moved. Changing the lead to state that it refers to actual terabit speeds collides with reality.

Petabit class devices section
In my understanding this section is wrong here and should be removed. The total device capacity is off-topic here, the article is about Ethernet PHYs. Any relation between these device and multiple-hundred Gbit Ethernet is missing. --Zac67 (talk) 11:03, 19 April 2018 (UTC)


 * I removed it as there was no mention to terabit Ethernet in the prose or the references. ~Kvng (talk) 12:58, 22 April 2018 (UTC)

Confusing for average reader
I wanted to know how many fiber strands each standard needs. And as an extra information how many wavelengths per fiber strand are used. But I don't understand this article.
 * 1) Is "media count" the same as fiber strands needed?
 * 2) According to the article 400GBASE-LR8 uses 1 media. Is this required fibers per direction? So to use it bidirectially you have to install a cable with a pair of fibers (2x the media?).
 * 3) And what are "lanes"? Is this the number of wavelengths per direction?

Kwinzman (talk) 23:53, 18 December 2019 (UTC)


 * Yes, "media count" means fiber strands per direction for optical PHYs. You need twice that in total for a link. The quotient of "lanes" and "media count" is the number of wavelengths/lambdas/colors in WDM. --Zac67 (talk) 08:49, 19 December 2019 (UTC)


 * Thanks! Makes sense now. Could you add that to the article somehow? Also with 400GBASE-SR4.2 it says 2 lanes right now. According to your explanation if I understood it correctly that is not correct and it should say "lanes: 2 per medium" instead? Thanks again! Kwinzman (talk) 10:27, 20 December 2019 (UTC)


 * We won't want detailed explanations in all PHY articles, I'll think about how to link that. We can hone details as in 400G-SR4.2 as soon as the specs are available; I don't have them yet (the standard is due pretty much now, specs to be released in spring). It's likely to use WDM: 2 wavelengths × 2 strands, 4 lanes in total. --Zac67 (talk) 10:34, 20 December 2019 (UTC)


 * 400GBASE-SR4.2 uses four pairs of MMF so eight fibres in total. There are two wavelengths per fibre, each transmitting in a different direction. This halves the number of fibres required compared to 400GBASE-SR8 — Preceding unsigned comment added by 94.175.212.46 (talk) 16:23, 20 February 2020 (UTC)


 * I have tried to make these columns less confusing over on the 100 Gbit table over on 100_Gigabit_Ethernet. I didn't realized this was also discussed over here.   please take a look at the changes I made. JeffMorriss (talk) 12:41, 27 April 2021 (UTC)
 * Great, looks much better now, especially the media counts! --Zac67 (talk) 07:36, 28 April 2021 (UTC)

Thanks for the feedback folks. OK, I've applied these changes to the 200G and 400G tables. Is it time to close/delete this issue? JeffMorriss (talk) 18:22, 28 April 2021 (UTC)

QSFPDD correct for 400GBASE-CR4?
For 400GBASE-CR4, the table lists "QSFPDD" in the "OFC or RFC" column. Is this really correct? QSFPDD provides 8 lanes in each direction (16 in total). However, only 4 lanes in each direction are needed (hence the "4" in "CR4"). Why would QSFP112 not be sufficient, then?

Also have a look at the "Form factors" in the bottom-right corner of 2020 Ethernet Roadmap, which shows "4-400 Gb/s" next to a QSFP (not DD) port.

Please let me know if I have missed something! 90.145.31.194 (talk) 08:27, 27 September 2021 (UTC)


 * Seems you're right. But 400G-CR4 isn't released yet, so it's hard to take a look. Feel free to contest or remove the QSFPDD connector right away or just wait until it's released (due in mid 2022). --Zac67 (talk) 20:08, 27 September 2021 (UTC)

800G table at the end of the page
Two things jumped out at me regarding the 800G table at the bottom of the page. It looks like this mostly is the result of copying / pasting the previous table:

(1) The top of the table (green background) says "Line rate: 8x 53.1250 GBd x2 = 425.0 GBd". This looks like it was copied from the previous table, and the number of lanes were changed from 4 to 8, but the total was not updated. It should say 850.0 GBd not 425.0 GBd.

(2) The reference given for this (number 37) cannot be correct and I suggest it be removed. Reference 37 is from mid-2017. There are zero mentions of 800G in this slide deck. Ideally another reference could be used, but for now this reference should be removed.

ChrisTracy (talk) 00:52, 16 May 2023 (UTC)


 * This might be a better reference: https://www.ieee802.org/3/df/public/22_07/healey_3df_01a_2207.pdf
 * There might be something even better under here: https://www.ieee802.org/3/df/index.html
 * I see some reverting back and forth on the line-rate. Maybe it should be Gbps rather than GBd so perhaps "Line rate: 8x 53.1250 GBd x2 = 850.0 Gbps" — Preceding unsigned comment added by 185.217.253.59 (talk) 16:26, 4 July 2023 (UTC)
 * The x2 for all PAM encodings which isn't present for NRZ certainly suggests some intent to calculate in Gbps rather than GBd. There's a bit of a mix of this across the tables. 2A02:8012:1295:FC:57B:D9C2:7294:F074 (talk) 14:59, 14 November 2023 (UTC)
 * Maybe we should just dump the = 212.5 GBd bit which is at least highly misleading (is only correct if you convert the PAM-4 symbols on the line to binary symbols somewhere). I wouldn't use Gbit/s here either as line code is usually represented in (G)Bd. --Zac67 (talk) 17:50, 14 November 2023 (UTC)

Lots of broken links
Most of the links in the intro for the Standards Development section are dead. IEEE appears to have removed the PDFs because of later revisions. The intro is probably the wrong place for these to be linked anyway. 2600:8800:1180:25:CDF3:8BA8:C69B:2F80 (talk) 18:40, 16 May 2024 (UTC)