Intel Readies the i225-V "Foxville" Low-cost 2.5 Gbps Ethernet PHY

[dinmaster]

2.5 lol should just jump to 10 as mainstream. Sounds like Intel wants to slow everything down and bring us up slowly.... like their cpu's. The big push for more will be when m.2 really takes off and everyone is using it. 1gbps is slow now, hdd in the past was the bottleneck but now I find the Ethernet the bottleneck.

 

[MazeFrame]

But they're also noisy, often fibre and huge. Most consumers wouldn't accept that, so if we're talking consumer adoption, that's a no go.
Obviously some of us are mad and don't care, or have a network rack or similar (I don't), but not everyone can have things like that at home.

40 to 50dB in a closet or in the attic is acceptable in my mind. Could also take a saw to the top cover and mod 120mm fans on there.
I'll admit that 2/3rds of the used data center switches are SFP+ instead of RJ45.

 

[Octopuss]

This is what the 2.5Gbps Killernic is based on.

That doesn't make sense.
How can a card be based on a chip that hasn't been released yet?

 

[TheLostSwede]

That doesn't make sense.
How can a card be based on a chip that hasn't been released yet?

Because Intel made a special deal? Wouldn't be the first time some company has done that.
Also a good way to get your gear out to "beta" test it on some consumers...
The Intel i225 series of network chips has also been "leaked" for quite some time, like in this news post about a couple of embedded boards from back in March...
It's just that the site in question, didn't really know what they were looking at.

Congatec takes a swig of Whiskey Lake in 3.5-inch and thin Mini-ITX SBCs

Congatec unveiled Linux-friendly, 3.5-inch “Conga-JC370” and thin Mini-ITX “Conga-IC370” SBCs with Intel’s 8th Gen 1.8GHz, quad-core Core i7-8565U and 2.1GHz, dual-core Core i3-8145U Mobile processors.

linuxgizmos.com

2.5 lol should just jump to 10 as mainstream. Sounds like Intel wants to slow everything down and bring us up slowly.... like their cpu's. The big push for more will be when m.2 really takes off and everyone is using it. 1gbps is slow now, hdd in the past was the bottleneck but now I find the Ethernet the bottleneck.

You might've missed the fact that Intel is among the last to launch a 2.5Gbps Ethernet controller.
Realtek, Marvell and Aquantia has had products our for some time already, alongside the Killer E3000 which is based on Intel hardware.

 

[Assimilator]

LOL @ everyone complaining about not going directly to 10Gbps or "it's another Intel conspiracy!!!!oneoneone"

It's basic economics. Intel's cheapest 1Gbe controller costs $1.72, its cheapest 2.5Gbe is "only" $4.06 or a mere 2.3x the price. 10Gbe will set you back $37.34 - only 9.1x as much as 2.5Gbe, and an insignificant 21.7x the cost of 1Gbe. Guess who the largest buyers of motherboards, and therefore ethernet controllers are? Corporates. Guess how many corporates actually need anything faster than 1Gbe? A handful. Guess what happens to motherboard manufacturers that tell the average big corporate that motherboards now cost $37 more (which is probably more than the rest of the motherboard costs) but have 10Gbe? They lose their supply contracts. Strangely, few manufacturers are willing to do this!

Not to mention the heat problem. 1Gbe controllers dissipate *half a watt* on a 40nm node, 2.5Gbe controllers almost 4x that, and 10Gbe controllers 75% more than that on a 28nm node. Or to put it another way, Intel's most efficient 10Gbe controller runs hotter (3.74W) than some of its 300-series chipsets (3W). You know another thing that manufacturers love, it's having to add a heatsink to the BOM and redesign their boards to fit a hot network chip that almost nobody wants.

Intel evidently thinks that the time (cost) is right to get the industry to 2.5Gbe. It is most certainly NOT the right time to go to 10Gbe, and won't be for many years. If you want to bellyache, bellyache at capitalism.

 

[TheLostSwede]

LOL @ everyone complaining about not going directly to 10Gbps or "it's another Intel conspiracy!!!!oneoneone"

It's basic economics. Intel's cheapest 1Gbe controller costs $1.72, its cheapest 2.5Gbe is "only" $4.06 or a mere 2.3x the price. 10Gbe will set you back $37.34 - only 9.1x as much as 2.5Gbe, and an insignificant 21.7x the cost of 1Gbe. Guess who the largest buyers of motherboards, and therefore ethernet controllers are? Corporates. Guess how many corporates actually need anything faster than 1Gbe? A handful. Guess what happens to motherboard manufacturers that tell the average big corporate that motherboards now cost $37 more (which is probably more than the rest of the motherboard costs) but have 10Gbe? They lose their supply contracts. Strangely, few manufacturers are willing to do this!

Not to mention the heat problem. 1Gbe controllers dissipate *half a watt* on a 40nm node, 2.5Gbe controllers almost 4x that, and 10Gbe controllers 75% more than that on a 28nm node. Or to put it another way, Intel's most efficient 10Gbe controller runs hotter (3.74W) than some of its 300-series chipsets (3W). You know another thing that manufacturers love, it's having to add a heatsink to the BOM and redesign their boards to fit a hot network chip that almost nobody wants.

Intel evidently thinks that the time (cost) is right to get the industry to 2.5Gbe. It is most certainly NOT the right time to go to 10Gbe, and won't be for many years. If you want to bellyache, bellyache at capitalism.

You got the pricing wrong, the i255-V is only $2.40, so not even 70 cents more, or around 40% more.

Beyond that, you're mostly correct.

 

[TechLurker]

Lisa save us! Mandate 5 and 10GbE across all of AMD's 570 and 550 motherboards! Put double 10 GbE on Threadripper boards!

 

[Makaveli]

No please don't...

After tall these years of sticking with 1Gb, you come now with 2.5Gb to make it mainstream ?

They should put 2.5Gb the new low end, 5.0Gb the new mainstream, and 10Gb for the new high-end (for both high-end mainstream platform and HEDT). And leave the 1Gb as the new very-low-end.

Do you have access to an ISP that offers more than 1Gbps packages?

That is FTTH only no one on cable or dsl has those kind of wan speeds.

There is no need to go to 5Gbps when very few ISP offer anything over 1Gbps.

If we are talking about sending data to a local client within your network like a NAS that is a difference story and I would agree with you in that use case.

I don't understand this.
I've got no issues with realtek .. a decade ago it was different though.
get 120 mb/sec on 1 gigabit realtek, intel 116 mb/sec o.0

Your example doesn't really mean anything in the grand scheme.

There is a reason intel Nics are considered the best. i'll leave it to you to research why.

Just consider how a new WiFi AX (WiFi 6) router costs upwards of $500, once you do that, $200-$400 for a managed multi gig 10GBase-T switch is pretty reasonable.

Not sure where you are looking but I picked up a Asus AX88U for $363 no where close to upwards of $500.

LOL @ everyone complaining about not going directly to 10Gbps or "it's another Intel conspiracy!!!!oneoneone"

It's basic economics. Intel's cheapest 1Gbe controller costs $1.72, its cheapest 2.5Gbe is "only" $4.06 or a mere 2.3x the price. 10Gbe will set you back $37.34 - only 9.1x as much as 2.5Gbe, and an insignificant 21.7x the cost of 1Gbe. Guess who the largest buyers of motherboards, and therefore ethernet controllers are? Corporates. Guess how many corporates actually need anything faster than 1Gbe? A handful. Guess what happens to motherboard manufacturers that tell the average big corporate that motherboards now cost $37 more (which is probably more than the rest of the motherboard costs) but have 10Gbe? They lose their supply contracts. Strangely, few manufacturers are willing to do this!

Not to mention the heat problem. 1Gbe controllers dissipate *half a watt* on a 40nm node, 2.5Gbe controllers almost 4x that, and 10Gbe controllers 75% more than that on a 28nm node. Or to put it another way, Intel's most efficient 10Gbe controller runs hotter (3.74W) than some of its 300-series chipsets (3W). You know another thing that manufacturers love, it's having to add a heatsink to the BOM and redesign their boards to fit a hot network chip that almost nobody wants.

Intel evidently thinks that the time (cost) is right to get the industry to 2.5Gbe. It is most certainly NOT the right time to go to 10Gbe, and won't be for many years. If you want to bellyache, bellyache at capitalism.

Finally someone with some enterprise experience clearly.

Couldn't have said it better myself.

 

[R-T-B]

Lisa save us! Mandate 5 and 10GbE across all of AMD's 570 and 550 motherboards! Put double 10 GbE on Threadripper boards!

Who'll make their NIC controllers? Intel?

 

[Patriot]

Who'll make their NIC controllers? Intel?

They had 10gbit IP in the original Zen1 CCX's Don't know if the current io/die has it in it or not... Each ccx basically has 2 optional ports...

 

[TheLostSwede]

They had 10gbit IP in the original Zen1 CCX's Don't know if the current io/die has it in it or not... Each ccx basically has 2 optional ports...

Still needs a PHY though.

 

[Patriot]

Still needs a PHY though.

True... https://www.microsemi.com/product-directory/ethernet-solutions/3581-physical-layer

 

[TheLostSwede]

True... https://www.microsemi.com/product-directory/ethernet-solutions/3581-physical-layer

Cheaper to by from someone like Aquantia, now part of Marvell... As they've already made the hardware.
They're about $30 or less in quantity, or were at least.

 

[Patriot]

Cheaper to by from someone like Aquantia, now part of Marvell... As they've already made the hardware.
They're about $30 or less in quantity, or were at least.

Microchip/microsemi ships 1B chips a year, I doubt a full 10gbit nic from aquantia would be cheaper than a phy from any other source.

 

[TheLostSwede]

Microchip/microsemi ships 1B chips a year, I doubt a full 10gbit nic from aquantia would be cheaper than a phy from any other source.

I wasn't talking about a NIC, I was talking about a PHY.

https://www.arrow.com/en/products/aqr107-b0-ig-y/aquantia-corp

Also, can you provide pricing for a 10Gbps PHY from Microsemi?
Their products don't exactly seem to be targeting PC applications though.

There aren't that many companies that makes 10Gbps PHYs.

 

[Patriot]

I wasn't talking about a NIC, I was talking about a PHY.

https://www.arrow.com/en/products/aqr107-b0-ig-y/aquantia-corp

Also, can you provide pricing for a 10Gbps PHY from Microsemi?
Their products don't exactly seem to be targeting PC applications though.

You can get a full aquantia nic in the 50-60 range. And nope, microchip does not target you or me they target the manufacturers so no clue on pricing but given the volume they sell... I have found thier chips on my Samsung ssds to amd gfx cards they make all the things. Marvell may try to undercut them being a smaller player but others have the volume to crush at will. I did notice that the 8 ports on the epyc 3000 said sfp+ which suggested a phy might already be included idk

 

[TheLostSwede]

You can get a full aquantia nic in the 50-60 range. And nope, microchip does not target you or me they target the manufacturers so no clue on pricing but given the volume they sell... I have found thier chips on my Samsung ssds to amd gfx cards they make all the things. Marvel may try to undercut them being a smaller player but others have the volume to crush at will. I did notice that the 8 ports on the epyc 3000 said sfp+ which suggested a phy might already be included idk

5Gbps, yes, the 10Gbps ones are around $80.

How do you know they don't target me? Do you know what I do for a living? I guess not...

Marvell recently bought Aquantia, most likely because Aquantia's parts are cheaper to make than Marvell's own parts. Marvel on the other hand makes comic books and movies...

I found some Microsemi pricing, and it's not cheap.

VSC8486YSN-04 Microchip / Microsemi | Mouser

VSC8486YSN-04 Microchip / Microsemi Ethernet ICs 10 GbE Phy WAN/LAN datasheet, inventory, & pricing.

www.mouser.com

SFP+ ports don't have a PHY, it's built into the transceiver.

 

[Patriot]

You do know that mobo makers don't go through mouser and digikey right?
I mean for protos sometimes they do but not for production volume.

And you linked $54 at 1 $44 at 100 500 are call me and nothing for large volumes.

As far as I can tell, you troll for a living. Not sure why you are attached to aquantia/marvell, when paging through block diagrams last night landed on a gigabyte board using a broadcom chip for it's phy.

 

[TheLostSwede]

You do know that mobo makers don't go through mouser and digikey right?
I mean for protos sometimes they do but not for production volume.

And you linked $54 at 1 $44 at 100 500 are call me and nothing for large volumes.

As far as I can tell, you troll for a living. Not sure why you are attached to aquantia/marvell, when paging through block diagrams last night landed on a gigabyte board using a broadcom chip for it's phy.

A lot of companies buy sample quantities from Arrow and Digikey and plenty of other distributors. Even production level quantities for smaller projects.
In fact, they're sometimes the official distributors of certain parts and there's no other channel, but you knew that, right?

It's comparable pricing, as the Aquantia pricing was for 1pcs as well. It might not be really viable for any sensible implementation, but it's an indication of price, which is what we were discussing, no? Or now you changed your mind? Got to love all the people changing the goal posts here in every single discussion when they're not winning their twisted argument.

What Gigabyte board? I don't see anything from Gigabyte. A Congatech board diagram that has optional 10Gbps support over some PCIe lanes is all I see, as well as a link to Microsemi. Maybe go back and check your posts? Or maybe you're the one trolling...

As for the relevance of Aquantia/Marvell, they're the only company that so far has offered a mostly consumer/prosumer focused 10Gbps solution and as everyone seems to want 10Gbps over 2.5Gbps in this thread, it's highly relevant. I don't see your pals at Microsemi doing things, nor does Intel, Broadcomm or anyone else that I'm aware of.

 

[Patriot]

A lot of companies buy sample quantities from Arrow and Digikey and plenty of other distributors. Even production level quantities for smaller projects.
In fact, they're sometimes the official distributors of certain parts and there's no other channel, but you knew that, right?

It's comparable pricing, as the Aquantia pricing was for 1pcs as well. It might not be really viable for any sensible implementation, but it's an indication of price, which is what we were discussing, no? Or now you changed your mind? Got to love all the people changing the goal posts here in every single discussion when they're not winning their twisted argument.

What link to what Gigabyte board? I don't see you linking to anything from Gigabyte. A Congatech board diagram that has optional 10Gbps support over some PCIe lanes is all I see, as well as a link to Microsemi. Maybe go back and check your posts? Or maybe you're the one trolling...

Where is your congatech board diagram link? No one is changing the goalpost but you.
I said I found it last night not linked it, you don't link everything you stumble across, it wasn't what I was looking for... there were a lot of rome block diagrams to wade through despite having searched for naples.

And I said prototype stage which is small volumes 100s not 1000s For some small niche products that could be production in which case the higher prices would work fine... as the component manufacturers aren't interested in such small quantities.

And fine, let's go back to the start.
They have the IP they just need the PHY, and you start getting an aquantia boner.
There are other vendors, no just aquantia for me bro. They are the only sensible solution.

But fine.

The epyc board I stumbled upon last night using a broadcom chip...found another using a full mellanox x4
I didn't bring it up because I have yet to find anyone using a phy, only whole chip solutions or SFP+ which doesn't require the phy.

I find the lack of phy only solutions intriguing and I really don't give a rats ass about your love for aquantia or how some vendors don't have enough volume to go direct to manufacturers.
Did you know that most companies don't buy servers direct either? Who cares.

Connect x4 for reference.

Best I can tell, AMD has the IP to do networking but because everything is over fabric and not preallocated no one has opted to use the AMD MAC and just use their own full chip solutions.

So the goalpost that got moved was, AMD doesn't have to use an Intel chip to do 10gbit, they have their own IP. You can make a thread to argue about mouser if you can find any other trolls to play with.

 

[TheLostSwede]

They had 10gbit IP in the original Zen1 CCX's Don't know if the current io/die has it in it or not... Each ccx basically has 2 optional ports...

This is not from Congatech? I'm sorry, but I'm really confused if it's not from Congatech...

And I'm not your "bro"...

I never said AMD needs Intel, please stop making stuff up, I simply pointed out that AMD needs a partner that can provide a PHY, as AMD doesn't make Ethernet PHY's.
Then you go on some mad rant...

 

[Patriot]

I never said AMD needs Intel, please stop making stuff up, I simply pointed out that AMD needs a partner that can provide a PHY, as AMD doesn't make Ethernet PHY's.
Then you go on some mad rant...

You told me to go look at the starting point and I did.
AMD has the IP, they need a phy, and you got hard for aquantia and went down some rabbit trail about why only aquantia is logical.
Your name is fitting.

 

[Tigerfox]

What I don't understand ist why it makes any sense to design an NIC or PHY especially für 2.5GbE. When NBASE-T was announced, everyone complained that the difference only lies in cabling and the tech needed was the same as 10GbE, only with lower frequencies due to cabling. Since that, too, was only necessary for distances above 55m, it would only be of benefit as a fallback mode for large buildings with old cabling, while homeusers had high chances of always achieving 10GbE.

When Aquantia presented their AQC107/108-NICs, everyone pointed out that it was the exact same NIC, just that the 108 was limited to PCIe x1 instead of x4 and thus to 5GbE. The Chip itself was identical in size. Aquantias reference layout AQN107/108 showed as much with the layout being nearly identical apart from PCIe. Thus far, only the AQN107 has been adopted by many companies. Since, Aquantia has presented the AQC111C/112C and AQC111U/112U, which are 5GbE and 2.5GbE-only-NICs for PCIe3.0x1 and USB3.0 respectively. The AQC111C has a much smaller footprint then the AQC108, so it seems there may be a benefit in being limited to 5GbE only, but that could just be because of just having one instead of four PCIe-Lanes. Again there is no difference between AQC111 and 112 apart from Speed.

Same with Broadcom, which by the was ony seems to offer Multigig PHYs, not NICs. Their 10GBASE-T/NBASE-T-offerings are of the exact same size as the corresponding 5GbE-PHY with the same amount of Ports.

From this I would assume singleport 5GbE-NICs could have a small advantage in price and usability compared to 10GbE-NICs by using only one instead of four PCIe-Lanes, but apart from that the requirements for chip and board should be the same. I don't see any benefit at all in being limited to 2.5GbE only. You could plug a 5GbE-NIC in every System that has at least a PCIe2.0x1-Port (standard since AMD AM2+ and Intel S1155) or USB3.0 (same or even earlier on Intel S1156), while 10GbE needs at least a PCIe2.0x4-Slot (x2 electrically would suffice) or USB3.1.

So, I can only guess there are some cost benefits I don't know of for Intel and Realtek to choose 2.5GbE for massmarket instead of 5GbE, or, as someone said, they want to milk this cow first and only offer 5GbE later on. But, since with PCIe4.0 one lane will be sufficient für 10GbE, the small price benefit for 5GbE will soon disappear.

But sadly we can't do much about it. If Intel and Realtek really offer their 2.5GbE-NICs so much cheaper than any 5GbE-offering and not much more expensive than current GbE-NICs, we should soon see Routers, Switches and Mainboards that offer only or atleast some 2.5GbE-Ports and don't cost much more than GbE-only-offerings, while faster speeds will remain too expensive for most users. That really is a shame because 2.5GbE will limit even one regular 3.5"-HDD, let alone SSD, while 5GbE wouldn't limit the former and even the later not so much.

 

[ditz]

My understanding based on some Marvell SoC datasheet... 2.5GbE is implemented like overclocking the 1Gb (R)GMII signals while the 5GbE is 10GbE with reduced clock.
So i think it make sense that 1GbE evolved to 2.5G instead of 5G since the base technology is quite different.