| 1 Jun 2025 |
 magic_rb | In reply to @magic_rb:matrix.redalder.org
The what now, im gonna get a fritzbox with a sfp pon from my provider soon, will try showing it into my bpi to see what happens :D I mean the pon, not the whole fritzbox :D | 11:48:20 |
 K900 | SFP is weird and A LOT | 11:48:46 |
 emily | I've ordered the 100m variant of https://www.luleey.com/product/10gbase-t-sfp-to-rj45-copper-module-30m-80m-100m/ and am just praying that it can:
- do 2.5GBASE-T into 2500BASE-X
- in a way that actually works with the BPI-R4 (new worry added today)
- and not mess up jumbo frames
- and preferably report temperature etc. stats
- and also do the thing where it passes through access to the PHY chip instead of hiding it behind pretending to be a fibre transceiver like apparently a lot of them do?
| 11:49:06 |
| emily | I've found that it's basically impossible to find clear statements from vendors on most of these | 11:49:21 |
| emily | so I'm probably resigned to just sorta buying and returning things until one of them does what I want | 11:49:36 |
| K900 | Like, there's modes where you have the SFP part just doing media conversion, and then there's modes where you have an entirely separate PHY that the device just talks to over SGMII or whatever | 11:49:46 |
| K900 | And everything in between | 11:49:51 |
| emily | by the way I found out that Marvell has a 5 nm 10GBASE-T PHY chip that consumes like < 1 W of power | 11:50:01 |
| emily | but NOBODY has put it in an SFP+ or switch for some reason | 11:50:07 |
| K900 | And then yes there's modes where the devices just fucking lie | 11:50:17 |
| emily | despite seeming like the holy grail of 10GBASE-T | 11:50:20 |
| K900 | Sometimes they both lie to each other just to agree on something | 11:50:35 |
| magic_rb | (Still talking about copper i take it) | 11:50:41 |
| emily | damn… kind of like people | 11:50:50 |
| magic_rb | (Im only interested in fiber for the bpi) | 11:50:53 |
| emily | yes this is about copper SFP+s | 11:51:01 |
| emily | well not even "copper". gah, what are you actually supposed to call Ethernet over 8P8C? | 11:51:13 |
| emily | BASE-T? | 11:51:17 |
| emily | looks like Wikipedia settles on "Ethernet over twisted pair". that's too long. | 11:51:40 |
| magic_rb | Ethernet over twisted pair, idfk | 11:51:43 |
| magic_rb | Ah | 11:51:48 |
| magic_rb | Guess im right lmao | 11:51:51 |
| magic_rb | I like being right :> | 11:52:02 |
| emily | I feel like "BASE-T" is reasonable. I'm too much of a pedant to say RJ45 but I need something short or I'll go insane. | 11:52:18 |
| emily | though "NBASE-T" refers only to 2.5GBASE-T and 5GBASE-T which, aaagh | 11:52:40 |
| emily | also yeah I'm really hoping for the latter here since it seems like it'll give the kernel more visibility/control | 11:53:01 |
| emily | like I won't have to worry about the SFP+ fucking up jumbo frames | 11:53:13 |
| magic_rb | Lol, btw, if 2.5g isnt real, then how do all those 2.5 motherboards do it? Like the copper version | 11:53:29 |
| emily | do you know what I'd look for in chip datasheets for that? or is it a decision by the transceiver manufacturer rather than the chip? | 11:53:31 |
| magic_rb | Or are just 2.5 sfps not real | 11:53:34 |
