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

No. How the protocol operates can't manipulate the laws of physics.
I think the lack of cheap 10GbE equipment has more to do with the laws of economies of scale and market demand for consumer-grade 10GbE than any physics. Most normal people have little to no use for anything beyond 1GbE yet. For many people, the only network wiring they have is the cable connecting their ISP gateway to the outside world. Among my immediate friends and family, I think I'm the only one using wired networking for everything that has an Ethernet port. My main PC is the only one with 2.5GbE and I have no plan to upgrade my 11ac router for 2.5+GbE while those routers still carry a steep premium.
 
  • Like
Reactions: Grobe
^^^^^^
Exactly.

In the consumer space, there is little need for above gigabit ethernet.
People aren't slinging a dozen 4k movies around the house on a daily basis.

And a LARGE segment of people have WiFi only.
 
I think the lack of cheap 10GbE equipment has more to do with the laws of economies of scale and market demand for consumer-grade 10GbE than any physics. Most normal people have little to no use for anything beyond 1GbE yet. For many people, the only network wiring they have is the cable connecting their ISP gateway to the outside world. Among my immediate friends and family, I think I'm the only one using wired networking for everything that has an Ethernet port. My main PC is the only one with 2.5GbE and I have no plan to upgrade my 11ac router for 2.5+GbE while those routers still carry a steep premium.
While I don't disagree the economies of scale will help cheapen the cost of getting everything up to 10GbE, my point is more if there's still a relatively high cost of high performance networking components, optimizing the protocol isn't going to change that.

It's like asking if DX12 would've made video cards cheaper.
 
  • Like
Reactions: bit_user
While I don't disagree the economies of scale will help cheapen the cost of getting everything up to 10GbE, my point is more if there's still a relatively high cost of high performance networking components, optimizing the protocol isn't going to change that.
Optimizations geared towards AI and HPC are unlikely to help the unwashed masses either. On a strictly per-port unit cost basis, I doubt 10GbE today costs a whole lot more than 1GbE did when it became mainstream ~16 years ago - we do have at least 10X more DSP processing power per dollar and watt than we had back then.
 
  • Like
Reactions: bit_user
We'll finally get cheap 10gbe ports for the pc?!?
As the article said, they're not looking to overhaul existing standards, but merely looking for opportunities to improve efficiency and scalability. They are probably concerned mostly with 100 Gbps and above.

I do think 10 Gbps will still become more mainstream, in time. It's been moving at a glacial pace, but faster Ethernet standards have become more common and affordable. 2.5 Gbps is finally becoming entrenched as the new standard for upper-end mainstream boards, and you can get 8-port, fanless, 2.5 Gbps switches for under $100.

And here's a 2.5 Gbps NIC for $13.70, shipped:


I bought one and tested it. The heatsink isn't necessary, but I think it's a nice touch. Most don't have one.
 
Last edited:
My main PC is the only one with 2.5GbE and I have no plan to upgrade my 11ac router for 2.5+GbE while those routers still carry a steep premium.
I just bought a DOCSIS 3.0 cable modem, which requires a 2.5 Gbps Ethernet port to reach maximum speed. I still have yet to replace my router, as I don't really need the extra speed (not to mention that I expect ustream bottlenecks will mean I rarely ever see it, in practice).

It is something of a milestone, I think. My first cable modem had a 10 Mbps port and I thought that was pretty cool (my first modem was 1200 baud, which was decidedly not very cool, as I think 14.4k modems were already the hot new thing, and you could almost read text as fast a 1200 baud modem would download it).
 
The demand for 2.5Gb Ethernet has shot up with the introduction of Wifi 6E, since 1Gb ethernet is not enough to feed a high performance AP. Nowadays 2.5Gb is becoming the new standard.

And 10Gbe is becoming cheaper, just now i read a review of a 8 port fully managed 10 Gb ethernet switch for less than 300$. Sadly it is imported from china and lacks any regulatory reviews, but it shows that prices are getting much better.
 
  • Like
Reactions: LuxZg and bit_user
And 10Gbe is becoming cheaper, just now i read a review of a 8 port fully managed 10 Gb ethernet switch for less than 300$. Sadly it is imported from china and lacks any regulatory reviews, but it shows that prices are getting much better.
Meanwhile, you can get an 8-ports 1GbE unmanaged switch from international brands with lifetime warranty for ~$30 which consume less than 5W total. Heat-wise, I cannot tell that my tiny 8-ports GbE Netgear switch is on, only way I know it has power is the blinking port LEDs.
 
1) Nvidia missing is sort of expected while still being quite rude from them, I'm certain they were invited

2) 2.5G is the new standard in everything from new MBOs, laptops, APs and NAS. But I've also seen that 10G switch review and while 10G is certainly NOT needed by most people, I'd still prefer if industry as a whole skipped minor steps and just went for 10G. I mean, cost isn't the stumbling issue at all, just look at prices of high end WiFi APs (6E/7) and they're pushing new WiFi standards constantly. At the same time they'll probably go from 1GbE to 2.5GbE to 3.75GbE to 5GbE if they can help it, milking each for 10+ years, putting the 10GbE at home in something like 2070's. We "don't need" it same as we don't need 8K TVs and similar, but somehow we're pushing everything else except the wired network speeds. That 245$ 10GbE switch (275$ for fully managed and PoE+ version!!) just proves that we could probably produce unmanaged non-PoE version of the switch for 100$ and still run with profit, using "good old" CAT6 cables for 99.999% of homes (55 meters is plenty). I've seen dual port 10GbE PCIe NICs for under 20$, so difference between 2.5GbE vs 10GbE single RJ45 port on MBO should be what... under 5$ for a motherboards that already sell for 300$+? If I can manage 300$ MBO I xan manage 305$ one as well if it brings 4x the Ethernet bandwidth, not to mention 2000$ GPUs and all.
 
2) 2.5G is the new standard in everything from new MBOs, laptops, APs and NAS. But I've also seen that 10G switch review and while 10G is certainly NOT needed by most people, I'd still prefer if industry as a whole skipped minor steps and just went for 10G.
Two compelling arguments for 2.5 Gbps are cabling and power.

With 10 Gbps over copper, you're much more limited in what type of cable you can use and how long it can be. For people with legacy Cat 5e in their walls, 2.5 Gbps can likely be used without issue. If their only choice was 10 Gbps, they might have to rip it out and replace with Cat 6 or 6a, even if they would've been satisfied with 2.5 Gbps of bandwidth.

In terms of power, the last specs I saw on 10 Gbps over twisted-pair indicate that it takes a lot more power to drive. I'm not sure how much of that is in the digital domain, where it could benefit from smaller process nodes, but I believe a significant amount of it is simply inherent in driving such a high-frequency analog signal over those distances. The amount of inductance seems like it could be nontrivial.

I've seen dual port 10GbE PCIe NICs for under 20$,
Certainly not new ones! I'd bet they were used SFP+ NICs.
 
  • Like
Reactions: TJ Hooker
Two compelling arguments for 2.5 Gbps are cabling and power.

With 10 Gbps over copper, you're much more limited in what type of cable you can use and how long it can be. For people with legacy Cat 5e in their walls, 2.5 Gbps can likely be used without issue. If their only choice was 10 Gbps, they might have to rip it out and replace with Cat 6 or 6a, even if they would've been satisfied with 2.5 Gbps of bandwidth.

In terms of power, the last specs I saw on 10 Gbps over twisted-pair indicate that it takes a lot more power to drive. I'm not sure how much of that is in the digital domain, where it could benefit from smaller process nodes, but I believe a significant amount of it is simply inherent in driving such a high-frequency analog signal over those distances. The amount of inductance seems like it could be nontrivial.

I do understand power concerns, but they're not that big. We're talking household, distances are short, bandwidth usage is mostly idle/low with relatively short peaks, specially at 10Gbps speeds.

Cables and choice are not an issue either. 1GbE adapters supports 10/100/1000, same way proper 10G port supports 1/2.5/5/10 speeds, and if you've bought laptop/PC with 10GbE port and your cabling or needs aren't that high, you can still run it at 1G or 2.5G. Power will drop accordingly as well.

Certainly not new ones! I'd bet they were used SFP+ NICs.

Those may have been used, IDK, but RJ45 copper. 10GbE copper NICs have been around for 15 years and I doubt they cost much in manufacturing. Take that 245$ fully managed switch that's 8 ports with power supply and case. Makes it 30$ per port, and if you drop the profit, case, power supply, and advanced features, it's like 10$ per port at most. 2.5G isn't free either so I still think 5$ difference per port between 2.5GbE and 10GbE on a MBO (in production). Not like I expect RoCE on a laptop.
 
1GbE adapters supports 10/100/1000, same way proper 10G port supports 1/2.5/5/10 speeds,
Nope. Many 10 Gbps NICs do not support "multi-gigabit" speeds. They jump straight from 1 Gbps to 10 Gbps, with nothing in between. It's generally newer chips that support those speeds, not least because 2.5 and 5 Gbps came along much later (2016) than the original 10 Gbps standard.

BTW, it sounded like you were arguing that the 2.5 and 5 Gbps standards shouldn't even have been created. If you're now talking about negotiating down to those speeds, then I guess we're agreed that they serve a useful purpose.

Those may have been used, IDK, but RJ45 copper.
They can't have been new. Even for a used NIC, that's a suspiciously low price. I can find one dual-SFP+ NIC for that much, but most are closer to $30 or more. Given that these are used, there's no telling how well they work or whether they have any problems. Furthermore, it seems that all of these cheap cards card are PCIe 2.0 x8, so you'd better have an x8 slot to spare. When you see things this cheap, there are usually good reasons why!

The cheapest I can find a 10 Gbps NIC with RJ-45 actually selling for, is $50 - and it's just 1 port (and also used):

You ought to check your facts. I'd bet you can probably find pricing data on 10 Gigabit MACs, if you're curious how much it would add to the COGS of a motherboard. Don't forget to apply the standard markup, too.

I just bought a motherboard with dual-10 Gigabit onboard and it cost me about $450. The only other thing at all special about it is the ASpeed AST2600 BMC on there. Based on the price difference between some boards that have a version with/without that BMC, I can say the markup isn't mainly from that.

Take that 245$ fully managed switch that's 8 ports with power supply and case.
Link?
 
Last edited:
He's probably talking about the Chinese 10GbE managed switch from a little while earlier. I like to keep my imported Chinese tech as close to dumb as bricks as possible, less likely to come with effective pre-loaded malware.

You can get UNmanaged 8x10GbE stuff from better-known brands like TRENDnet, D-Link and TP-Link starting around $250-300, not exactly consumer-friendly pricing when 8x1GbE ones are ~$25.
 
  • Like
Reactions: bit_user
He's probably talking about the Chinese 10GbE managed switch from a little while earlier.
I asked because I don't want to presume what @LuxZg meant. If the price is from AliExpress, then I intend to point out that the approach of using them to estimate costs and applying those same costs to motherboards that aren't grey-market imports is invalid.

I like to keep my imported Chinese tech as close to dumb as bricks as possible, less likely to come with effective pre-loaded malware.
Same. Years ago, I decided unmanaged switches were about the only thing I felt safe to buy from Chinese brands. Especially routers, because I tend to keep up-to-date on my firmware for them, and you only need a backdoor or malware to be in one of the firmware updates to get hacked.

I ended up getting a managed multi-gigabit switch by Netgear. I have a Netgear router now, too. It's based on a Qualcomm Snapdragon of some sort.

You can get UNmanaged 8x10GbE stuff from better-known brands like TRENDnet, D-Link and TP-Link starting around $250-300,
Yes, I think switch pricing has been one of the factors holding back consumer adoption of 10 Gigabit. I paid $290 for my Netgear switch, although it has just 2x 10 gigabit ports. It complements those with 2x 5 gigabit ports, 4x 2.5 gigabit ones, and 4x 1 gigabit ports. Believe it or not, this works very well for me. I have only two machines that really benefit from being at 10 gigabit. If I ever get more, I can link in a 5-port unmanaged 10 gig switch to add 4 more. My backup/multi-media RAID is disk-based, so 5 gigabits is plenty for it. I have two other devices connected at 2.5 gigabits and plan to add a 3rd.

not exactly consumer-friendly pricing when 8x1GbE ones are ~$25.
I remember when 5-port, 1 Gigabit switches hit $100. That's around when gigabit adoption started taking off.
 
Last edited:
Yes, I think switch pricing has been one of the factors holding back consumer adoption of 10 Gigabit.
The other is need.

How many people do enough internal networking and data movement to warrant anything beyond the gigabit LAN?

Even if it were a mere $5 per 'device', I'd have to think real hard.
Switches, NICs, cabling beyond Cat5e, etc, etc....


Yes, this will change. But currently? I don't see it.
 
The other is need.

How many people do enough internal networking and data movement to warrant anything beyond the gigabit LAN?
Back in 2011, my backup RAID already ran at 330 MB/s, which could easily justify 2.5 or 5 gigabit. With people RAIDing SSDs, even 10 gigabit is justifiable.

If you copy around or load/save VM snapshots, for instance, that's where you'd really feel these higher network speeds. Another is NFS/CIFS-mounting your home directory, as I do. It lets me transparently share files between machines, because the storage is centralized.

The thing is: most people who need that level of speed know who they are. And for those of us who do, it's nice to have affordable options.

BTW, Toms doesn't really cater to the homelab set. If you try and poo poo 10 gigabit over on ServeTheHome, you'll get shouted down rather quickly.
 
  • Like
Reactions: TJ Hooker
Back in 2011, my backup RAID already ran at 330 MB/s, which could easily justify 2.5 or 5 gigabit. With people RAIDing SSDs, even 10 gigabit is justifiable.

If you copy around or load/save VM snapshots, for instance, that's where you'd really feel these higher network speeds. Another is NFS/CIFS-mounting your home directory, as I do. It lets me transparently share files between machines, because the storage is centralized.

The thing is: most people who need that level of speed know who they are. And for those of us who do, it's nice to have affordable options.

BTW, Toms doesn't really cater to the homelab set. If you try and poo poo 10 gigabit over on ServeTheHome, you'll get shouted down rather quickly.
I get all that.

And that is, currently, a very niche space.


RAID 0 SSDs? Ok, yeah...you go with that.
 
Even if it were a mere $5 per 'device', I'd have to think real hard.
Switches, NICs, cabling beyond Cat5e, etc, etc....
As long as your equipment that needs 10G are close enough to the switch, you can run direct cables without much issue. If the run is short enough, you may even be able to continue using your cat5e. Once you start having to re-run in-wall wiring because it is too long or old, it becomes exceedingly painful and that is how 1GBase-TX died despite being much cheaper hardware-wise than 1GBsae-T at the time.

Standards that require new wiring to work properly over meaningful distances start with a huge handicap, which is another reason we got 2.5G and 5G: accommodate different wiring run lengths and qualities besides 10G-or-bust.
 
I get all that.

And that is, currently, a very niche space.
What's your point, really? Are you saying this just because you can? Or are you concerned anyone reading this thread might run out and buy 10 Gigabit hardware they don't need?

RAID 0 SSDs? Ok, yeah...you go with that.
I never said anything about RAID 0. I don't have any RAID 0's. I'm feeling like this exchange isn't productive.