News Linux Foundation Creates Ultra Ethernet Consortium with Cisco, Microsoft, AMD and More

[USAFRet]

I never said anything about RAID 0. I don't have any RAID 0's. I'm feeling like this exchange isn't productive.

From you...
"With people RAIDing SSDs, even 10 gigabit is justifiable."

Unless you meant some other RAID type.

But whatever. I'm out.

 

[bit_user]

From you...
"With people RAIDing SSDs, even 10 gigabit is justifiable."

Unless you meant some other RAID type.

What I had in mind was that people are starting to put SSDs in a NAS (presumably RAID-5). There are even SSDs marketed for NAS usage, and we've seen NAS boxes with 10 gigabit interfaces on them for years.

I didn't advocate for this practice. I'm just pointing out that it's a reality.

 

[USAFRet]

What I had in mind was that people are starting to put SSDs in a NAS (presumably RAID-5). There are even SSDs marketed for NAS usage, and we've seen NAS boxes with 10 gigabit interfaces on them for years.

I didn't advocate for this practice. I'm just pointing out that it's a reality.

I have an SSD in my NAS.
Only the one, though.

Accessing data in the NAS, be it on the SATA III SSD or any of the HDDs (single or RAID 5), is zero difference over the gigabit LAN.
Would it be "faster" on a 10Gbe? Yeah, maybe.
But not nearly faster enough for my use case, in my environment.

 

[LuxZg]

Nope. Many 10 Gbps NICs do not support "multi-gigabit" speeds. .....

<cut>
...
Link?

Sorry, been away a little.
Anyway, you're arguing different things.

I am saying that in 07/2023 technology and manufacturing have come far enough that we could just skip 2.5 (and possibly 5) iterations as they are late to the game, and just look like milking the consumer for future upgrades.

Costs in manufacturing 2.5G vs 5G vs 10G single port consumer MBO certainly aren't THAT high not to at least give it a push. We're pushing Thunderbolt 4 and WiFi 7 on higher end motherboards and these cost more to implement than the difference between 2.5G vs 10G.

You're also talking about cables, and I gave you perfectly valid example of a solution in autonegotiation of a lower speed. You reply with old NICs not supporting it. I'm talking about adding it to new MBO. If I get new MBO I don't need old NIC so I don't want to talk about that at all. I'm talking about going from 1G straight to 10G when buying contemporary hardware.

You're also talking expenses from current marketing view. Yes, ASUS will sell you a 2x10G "pro creator whatever" board for a 200$ markup, because even if it was same with 1x1G but painted pink they'd milk you. That does NOT say that cost per port is 100$.

I've pointed to that switch to show that per port price isn't all that high. I don't want to comment on "Chinese security issues" (which are exaggerated to the Moon and back), but if it cost Chinese manufacturer 245$ (with markup) then Taiwanese one can do so as well.

Link:

Finally A Cheap 8-Port 10Gbase-T Managed PoE Switch the Hasivo S1100WP-8XGT-SE

We finally found a cheap, managed, 8-port 10Gbase-T switch with PoE in the Hasivo S1100WP-8XGT-SE at only $278, a fraction of the competition

www.servethehome.com

I can reply to all those speed/cable/power/whatever concerns of yours by pointing back at other tech.... USBs, WiFi, and so on. "Most people don't have/need ..." - yes, same with WiFi 7 or Thunderbolt 4. "This is so expensive ..." - yes, same as USB/Thunderbolt 4 and WiFi 7. "You need new routers and switches..." - yes, and you need new hubs, docks, and APs for either WiFi7/USB4. "But this or that device does not support such high speeds .." - yes, and not all mobile devices do WiFi7 nor all your USBs go to USB4, you have to start somewhere, and that's what autonegotiation and backwards compatibility are for (get hub/switch that "does them all" and let it negotiate the correct speed for your "legacy" devices). "I would think very hard if I'd buy a 5$ more expensive...." - yet people buy 1200$ Samsung or Apple phone then immediately upgrade APs to WiFi 7 (500-1000$) and then need to upgrade link from AP to other endpoint to 2.5G (or 10G) just so they can claim faster Instagram loading times.

And so on.

We've moved from "WiFi 5" (ac) to WiFi 7 in 10 years, jumping from "WiFi4" ( n ) in under 15 years. Some people are still using b/g devices. Jump from "4" to "7" in 15 years provided jump from 600Mbps to 46Gbps. And here we argue if we should switch "straight" to 10G on wire, when we've used gigabit Ethernet since 1999 (almost 25 years soon!).

In same 25 years we've gone from Pentium III to Gen 13. That's roughly 300x performance. We've gone from 12Mbps USB 1.1 to USB 4 at 80Gbps, so some 6000x faster. 3G from 2001 and 8Mbps vs 5G today... Etc...

But noooo, we'd rather have a lightning struck us than admit that it's time to move away from Gigabit Ethernet and that we deserve a jump higher than 2.5x in 25 years.

Excuse my typos, I caught some, but I'm typing on a phone...

 

[bit_user]

I'm talking about adding it to new MBO. If I get new MBO I don't need old NIC so I don't want to talk about that at all. I'm talking about going from 1G straight to 10G when buying contemporary hardware.

Okay, let's agree to agree. I agree that it would be nice if Mobos would just add 10 Gigabit MACs, which are capable of down-negotiating to "multi-gigabit" speeds. With regard to talking about current hardware, beware that there's still a Broadcom Ethernet controller being shipped in new products that does not support 2.5 Gbps and 5 Gbps, however.

What I wish we knew was the actual difference in parts costs. If you feel like researching that, it would further the discussion. I'm not terribly interested in trying to reverse-engineer parts cost based on some switch you can only buy on AliExpress - many of those products don't have the proper regulatory approval for resale in North America or Europe and could have sourced parts at cut-rate prices intended only for the domestic Chinese market.

Instead, how about looking up the part numbers and pricing them in 10k quantities from an electronics wholesaler? You don't have to, but I think that's what it would take to meaningfully further the discussion. Otherwise, I'll just say that we agree in our hopes and vision.

I'll go first. Here's Intel's X550, a popular controller found on many 10 Gigabit motherboards and NICs. $32.20 in qty. 600

​

https://www.mouser.com/ProductDetail/MaxLinear/GRX550A3BC200?qs=pBJMDPsKWf3BMizkInaI0g%3D%3D

​

Maybe @InvalidError can tell us what the heuristic is for estimating the final MSRP, when adding that much to the COGS of something like a motherboard.

If you want to help, maybe price out the Aquantia/Marvell controller? I believe it's called the AQN-107, but I can't find it on either of the sites I checked.

And here we argue if we should switch "straight" to 10G on wire, when we've used gigabit Ethernet since 1999 (almost 25 years soon!).

1 gigabit Ethernet wasn't mainstream until the early 2000's. So, I'd say 20 years. When I bought a motherboard in 2004, on-board Gigabit was still a premium feature. And, as I also mentioned, a cheap 5-port gigabit switch cost about $100 in early 2003 - I remember that I was surprised to see that!

In same 25 years we've gone from Pentium III to Gen 13. That's roughly 300x performance.

Okay, how about comparing the single-threaded performance of a Pentium 4 in 2003 vs. Raptor Lake in 2023? Because that's what we're talking about: serial communications, not parallel.

We've gone from 12Mbps USB 1.1 to USB 4 at 80Gbps, so some 6000x faster.

Again, if we start counting from when gigabit went mainstream, then you'd be comparing vs. USB 2.0 and its bitrate of 480 Mbps. Also, it's no fair comparing against 80 Gbps when you can't run a passive cable nearly as far at that rate. You should rather compare to 10 or 20 Gbps.

 

[LuxZg]

Okay, let's agree to agree. I agree that it would be nice if Mobos would just add 10 Gigabit MACs,

...

Again, if we start counting from when gigabit went mainstream, then you'd be comparing vs. USB 2.0 and its bitrate of 480 Mbps. Also, it's no fair comparing against 80 Gbps when you can't run a passive cable nearly as far at that rate. You should rather compare to 10 or 20 Gbps.

Yeah, I'm in no mood to research that far ;D glad we agree (mostly) about wishes. I'll even accept that 480Mbps vs minimal 10Gbps... That's still 20x right? So we should aim for 25G wired instead 2.5G

 

[InvalidError]

Yeah, I'm in no mood to research that far ;D glad we agree (mostly) about wishes. I'll even accept that 480Mbps vs minimal 10Gbps... That's still 20x right? So we should aim for 25G wired instead 2.5G

How many people would have any sort of meaningful use for 10+G networking at home? Probably less than 0.01%. That is why you aren't going to see 10G in the mainstream until the cost of parts almost reaches parity with 2.5GbE.

BTW, USB 10-40G has an official maximum passive cable length of only 1m vs 100m for Ethernet when using the prescribed cabling. Pushing high bandwidth over longer cables is intrinsically more challenging.