Wednesday, October 2nd 2019
Intel Readies the i225-V "Foxville" Low-cost 2.5 Gbps Ethernet PHY
Intel is readying the i225-V "Foxville," its new generation of low-cost Ethernet PHY controllers for client-segment motherboards and notebooks. With it, the company will be mainstreaming 2.5 Gbps as the client-segment wired-networking standard, after nearly 15 years of 1 GbE dominance. The i225-V is expected to feature in the upcoming wave of socket LGA2066 motherboards for Intel's "Cascade Lake-X" HEDT processors, followed by the company's 400-series chipset that launches alongside the "Comet Lake-S" MSDT processors. The i225-V isn't the first of its kind, with the likes of Realtek and Broadcom having already launched 2.5 GbE PHYs. The Intel chip, however, is expected to mainstream the standard as it's currently the most popular GbE PHY brand with the success of the i219-V and i218-V.
Much like the i219-V, the i225-V is a low-cost PHY that relies on PCH-based Ethernet MAC and its proprietary PCIe-based bus that runs at half the data-rate of PCIe. This is precisely why the i219-V doesn't feature on AMD motherboards, but rather its pricier sibling, the i211-AT, which comes with an integrated MAC and a standard PCIe interface. Both chips are known to offer identical throughput performance, however, the i211-AT edges ahead with some features such as TCP segmentation, direct cache access, etc. The i219-V sells for as little as $1.5 per chip in high-volume reels to motherboard manufacturers, and the i225-V is expected to be priced roughly similar. In contrast, the i211-AT goes for almost $3.25 a pop. Intel is yet to publish documentation that details software features of the i225-V, but the Linux community is already on the job at developing drivers. 2.5 GbE uses existing Cat5E/Cat6 cabling requirements as 1 GbE, and hence has a better chance at mainstreaming compared to 10 GbE, which has been around for a decade, with little success in the client segment.
Sources:
Phoronix, siuol11 (Reddit)
Much like the i219-V, the i225-V is a low-cost PHY that relies on PCH-based Ethernet MAC and its proprietary PCIe-based bus that runs at half the data-rate of PCIe. This is precisely why the i219-V doesn't feature on AMD motherboards, but rather its pricier sibling, the i211-AT, which comes with an integrated MAC and a standard PCIe interface. Both chips are known to offer identical throughput performance, however, the i211-AT edges ahead with some features such as TCP segmentation, direct cache access, etc. The i219-V sells for as little as $1.5 per chip in high-volume reels to motherboard manufacturers, and the i225-V is expected to be priced roughly similar. In contrast, the i211-AT goes for almost $3.25 a pop. Intel is yet to publish documentation that details software features of the i225-V, but the Linux community is already on the job at developing drivers. 2.5 GbE uses existing Cat5E/Cat6 cabling requirements as 1 GbE, and hence has a better chance at mainstreaming compared to 10 GbE, which has been around for a decade, with little success in the client segment.
48 Comments on Intel Readies the i225-V "Foxville" Low-cost 2.5 Gbps Ethernet PHY
I'll admit that 2/3rds of the used data center switches are SFP+ instead of RJ45.
How can a card be based on a chip that hasn't been released yet?
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.
linuxgizmos.com/congatec-takes-a-swig-of-whiskey-lake-in-3-5-inch-and-thin-mini-itx-sbcs/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.
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.
Beyond that, you're mostly correct.
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.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.Not sure where you are looking but I picked up a Asus AX88U for $363 no where close to upwards of $500.Finally someone with some enterprise experience clearly.
Couldn't have said it better myself.
They're about $30 or less in quantity, or were at least.
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.
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.
www.mouser.com/ProductDetail/Microchip-Microsemi/VSC8486YSN-04?qs=sGAEpiMZZMuXKgZRMPEonYHObE10bvNc6RJWzz1V6%252Bw%3D
SFP+ ports don't have a PHY, it's built into the transceiver.
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.
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.
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.
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... :confused:
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.
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.
So i think it make sense that 1GbE evolved to 2.5G instead of 5G since the base technology is quite different.