Why doesn't Intel and AMD Integrate Gigabit Ethernet?

[24601]

Why doesn't Intel and AMD Integrate Gigabit Ethernet in either their CPU/Package/Northbridge/Southbridge?

There must be a good reason that everyone has motherboards with Realtek Gigabit Ethernet still to this day.

 

[Piroko]

Maybe power consumption of the Nic (about 1W) or royalties perhaps.

 

[Insert_Nickname]

Its simple. When manufactures buy realtek sound chips, they get a bundle deal on realtek NICs. You can still find some mainboards with Intel and Broadcom NICs if you look hard enough.

For the record, almost all Intel boards have Intels own NICs.

 

[24601]

Its simple. When manufactures buy realtek sound chips, they get a bundle deal on realtek NICs. You can still find some mainboards with Intel and Broadcom NICs if you look hard enough.

For the record, almost all Intel boards have Intels own NICs.

I'm not talking about dedicated (Dedicated Integrated ) NICs. I'm talking about Integrated into an already existing required component like I listed in my OP.

Gigabit Ethernet takes board-space, and I hear every day about how less PCB layers are cheaper (Why ASROCK boards are extremely cheaper or at least thats what people tell me).

Would also assume it would cut another manufacturer out of the process to further consolidate what goes into an Intel or AMD motherboard.

I understand that its better to isolate the sound into a separate IC for interference reasons, but don't see any technical limitations in fully integrating the NIC into the aforementioned components.

 

[BuffaloChuck]

I'm not sure about PCP space-consumption is a reason - the NIC-on-a-chip is pretty small these days, smaller than the space that the NIC-plug-in takes up on the board's edge. Still, integrating that NIC-chip would free up SOME space (not that boardmakers have a history of Best Use of Available Space - still, miracles COULD happen, I suppose).

I think AMD, Via and all non-Intel board producers would be losing a partner if they disintegrated NIC-providers. Even if Broadcom and Realtec aren't producing motherboards or CPUs, they're still providing Alternates-to-Intel engineering, employment, imagination and funding. Having more of those is probably better than giving Intel a larger share.

I'm not sure these other manufacturers truly consider THAT as a reason to Integrate/Not Integrate NIC services. Venture capitalists might, though.

 

[dealcorn]

Intel will build 4 Gigabit Ethernet ports into its Avoton chip (Xeon Atom @ 22nm) which is due prior to year end. There is talk that some models will also support an integrated optical port for server room duty.

 

[ShintaiDK]

It would be nice to get rid of all those 3rd party NICs. But Its all about saving half a dollar. And the OEMs simply demand it this way for now.

Personally I would cry crocodile tears if we "lost" Broadcom, Marvell and Realtek in the PC network segment.

 

[BrightCandle]

Besides Intel would cannibalise its own nic business where it gets to sell its superior technology at a premium.

 

[24601]

Besides Intel would cannibalise its own nic business where it gets to sell its superior technology at a premium.

I doubt it.

If they just integrate 1 Gbit Ethernet port they wouldn't cut into their server board profit margin since they could just price what the Intel NIC would cost into the cost of the component they are integrating it into.

In the consumer space almost no one uses the more expensive Intel NIC anyways, so its pure profit.

Also, on any server board that needed more than 1 Gigabit Ethernet port they would have to purchase whatever NICs they would have done so anyway.

Sounds like a win/win/win situation for Intel at the very least.

And they could always just fuse off (Or turn off in drivers, since people who run servers aren't usually into modding them ) the NIC on the wherever the NIC is integrated into the server.

 

[jpiniero]

Still think you will see more things get integrated into the die/package, to the point where there might not be much else needed.

 

[Vic Vega]

I don't think it makes sense to add this functionality.

 

[Yuriman]

Some machines don't even use NICs (ex MBA).

 

[Insert_Nickname]

Personally I would cry crocodile tears if we "lost" Broadcom, Marvell and Realtek in the PC network segment.

Me too... ()

 

[Arachnotronic]

Why doesn't Intel and AMD Integrate Gigabit Ethernet in either their CPU/Package/Northbridge/Southbridge?

There must be a good reason that everyone has motherboards with Realtek Gigabit Ethernet still to this day.

Intel will be integrating GbE into its "Avoton" (8 core Atom) SoC for Microservers.

 

[Cerb]

It would be nice to get rid of all those 3rd party NICs. But Its all about saving half a dollar. And the OEMs simply demand it this way for now.

Personally I would cry crocodile tears if we "lost" Broadcom, Marvell and Realtek in the PC network segment.

Leaving us with drop-the-connection-randomly Atheros?!

But, yeah, if an Intel CPU guaranteed an Intel Ethernet NIC, and better yet, also an Intel WiFi NIC...oh, please don't throw me into that briar patch! :biggrin:

 

[cheez]

I think CPU failure rate will increase with too much garbage running in the core. Also you will run into compatiblity issues with existing motherboards.

There is an old saying "Keep it Simple".


cheez

 

[Homeles]

I think CPU failure rate will increase with too much garbage running in the core. Also you will run into compatiblity issues with existing motherboards.

There is an old saying "Keep it Simple".


cheez

Even with all of the stuff they pack onto CPUs these days, failure rates are quite low. Also, Intel would definitely never introduce this on an old socket. AMD on the other hand is too scared to do anything that would break socket compatibility.

 

[ShintaiDK]

I think CPU failure rate will increase with too much garbage running in the core. Also you will run into compatiblity issues with existing motherboards.

There is an old saying "Keep it Simple".


cheez

Failure rate not caused by user right now on CPUs is something like 0.00001%

 

[lagokc]

Failure rate not caused by user right now on CPUs is something like 0.00001%

I think he was talking about yields. The larger you make the chip, the greater the chance that part of the chip is on a defective part of the wafer and as a result the smaller percentage of the wafer ends up in fully functional chips. I could be wrong though, and knowing Intel they'd just sell the chips with defective ethernet controllers as XXXXNE (no ethernet) edition and knock a couple $ off the retail price or sell them to OEMs to use in ultrabooks intended to have wireless only.

 

[ShintaiDK]

I think he was talking about yields. The larger you make the chip, the greater the chance that part of the chip is on a defective part of the wafer and as a result the smaller percentage of the wafer ends up in fully functional chips. I could be wrong though, and knowing Intel they'd just sell the chips with defective ethernet controllers as XXXXNE (no ethernet) edition and knock a couple $ off the retail price or sell them to OEMs to use in ultrabooks intended to have wireless only.

Yield is not an issue.

Also ethernet controller would be on the chipset, not the CPU. And the chipset is shrinked from 65nm to 32nm between 7 and 8 series.

 

[SocketF]

I once red, that the physical interface (PHY) cannot be shrunk below 65nm. So you could only integrate the network logic in a contemporary sub 45nm design. The PHY has to be external and it is connected by a modified PCIe x1 interface.

However, if you have to use a PCIe x1 interface anyways, then you can also connect to a damn cheap Realtek networkchip, which is produced in the good old 65nm at nearly no costs ...

If I remember correctly then some AMD SBs had an integrated Broadcom network logic, but next to nobody used it, because Broadcoms PHY-chip is more expensive than a Realtek chip. Same story for some intel chipsets.

It is also an logistic problem. If you are a mainboard manufacturer then it is cheapest to just store 1 kind of network chips that fits everywhere. Thus you can buy at high volumes and low costs. Buying Broadcom and/or Intel PHYs for a few kinds of special mainboards doesnt pay off.

 

[ShintaiDK]

I once red, that the physical interface (PHY) cannot be shrunk below 65nm. So you could only integrate the network logic in a contemporary sub 45nm design. The PHY has to be external and it is connected by a modified PCIe x1 interface.

However, if you have to use a PCIe x1 interface anyways, then you can also connect to a damn cheap Realtek networkchip, which is produced in the good old 65nm at nearly no costs ...

If I remember correctly then some AMD SBs had an integrated Broadcom network logic, but next to nobody used it, because Broadcoms PHY-chip is more expensive than a Realtek chip. Same story for some intel chipsets.

It is also an logistic problem. If you are a mainboard manufacturer then it is cheapest to just store 1 kind of network chips that fits everywhere. Thus you can buy at high volumes and low costs. Buying Broadcom and/or Intel PHYs for a few kinds of special mainboards doesnt pay off.

Seems Intel sells 40nm ethernet controllers. So I dont think that theory holds.

http://ark.intel.com/products/70831/Intel-Ethernet-Connection-I217-V
http://ark.intel.com/products/60019/Intel-Ethernet-Connection-I217-LM
http://ark.intel.com/products/60020/Intel-Ethernet-Controller-X540-AT2
http://ark.intel.com/products/60021/Intel-Ethernet-Controller-X540-BT2

 

[Khato]

Seems Intel sells 40nm ethernet controllers. So I dont think that theory holds.

It actually is correct - just because you're running on a 40nm process doesn't mean that you're benefiting from it. Unfortunately such is the case with many analog ICs, there's effectively no reason to use a smaller process since you're just making huge transistors anyway. Far more economical to use an older 'cheap' process for the task.

It'll be interesting to see how the low power I/O transistors of Intel's 22nm SoC process play out in this respect. It may be the case that with the finfet properties these types of transistors can finally get at least some process scaling again.

 

[Vinwiesel]

I once red, that the physical interface (PHY) cannot be shrunk below 65nm.

Sounds plausible. Many analog circuits don't shrink like a digital processor does.

 

[Wall Street]

Also keep in mind that the machines used to make these are fully depreciated, are not really useful for any high performance chips and have limited demand. The cost of making these ethernet controllers almost literally the cost of the silicon. I bet if some of the fabs making these were used on these parts they would just have more idle time since these nodes are no longer as useful for CPUs, chipsets and phones.