Intel Broadwell: goodbye chipset?

First there was the Northbridge+Southbridge couple, then there was the Platform Controller Hub (PCH) and soon there will be only the CPU to rule all the busses on the motherboard? According to the anticipations from ComputerBase.de, a couple of years from now Intel could remove yet another component of the motherboard chipset integrating it in the CPU package itself.

The complete integration of busses management in the CPU is the final iteration of a process already started by Chipzilla years ago with the introduction (on “5 Series” chipsets) of the aforementioned PCH, the chip on the motherboard designed to control some of the communications previously assigned to the Northbridge (PCI Express lanes) and the Southbridge (PCI slots, IDE/SATA, USB, Ethernet etc.) after having integrated the memory controller on the CPU.

According to the slides shown by ComputerBase.de, the “Broadwell” microarchitecture will push the integration process to its limits by merging the PCH with the CPU: Broadwell will be produced on a 14 nanometers node and will be based on a “Multi Chip Module” design – ie a single package will contain both the CPU+GPU core and the chipset controller.

Intel will seemingly market the Broadwell architecture as a “System-on-a-chip platform”, a definition thus far used for highly integrated devices like smartphones and other MIDs (Mobile Internet Device) but not for complex beasts full of boards, slots and components like desktop or server computers are.

Report a problem with article
Previous Story

Developer FourBros Studios show huge WP7 growth

Next Story

Five technologies to look for in your next smartphone

35 Comments

Commenting is disabled on this article.

All I can see happening in 10 years is no more motherboards, peripherals. anything.

you will buy a CPU that does everything. including power itself...

Now thats a bit extreme but that's how I see it heading. integrated motherboards as one unit with slots for "one unit" peripherals

Auzeras said,
All I can see happening in 10 years is no more motherboards, peripherals. anything.

you will buy a CPU that does everything. including power itself...

Now thats a bit extreme but that's how I see it heading. integrated motherboards as one unit with slots for "one unit" peripherals

It does seem to be heading that way. I still have my IBM PC/XT which had almost nothing built in. Need a math coprocessor? Seperate chip. Floppy drive? Controller card. Serial ports, parallel, video, hard disk controller, sound, everything was handled through expansion cards. Even the RAM was seperate chips instead of modules. Now they can put that entire system on a single chip.

eilegz said,
so its the equivalent of amd fusion?

The fusion chips are actually great overclockers, mine for example, goes from 1.6 up to 3ghz no problem.

Breakthrough said,

Can anyone explain to me how a chip drawing less power runs hotter?
Smaller process, more leakage on the chip. Which translates into heat.

alwaysonacoffebreak said,

Its no news that Intel is having problems with the 22nm, can't blame them on that, besides IB is an Tick on their roadmap so You shouldn't wait a big improvement from SB.

For sure, 22nm is already very close to the physical limits (16nm) before electrons are simply too big to work, regardless of what they do with it.

Still, if they are having problems going down to 22nm maybe they should be looking at alternative ways of getting more out of less. We already went the route of multiple die on one silica for the cpus to save the ghz being cranked up... what would we be at now, 8ghz + with 150c temps?

sagum said,

For sure, 22nm is already very close to the physical limits (16nm) before electrons are simply too big to work, regardless of what they do with it.

Still, if they are having problems going down to 22nm maybe they should be looking at alternative ways of getting more out of less. We already went the route of multiple die on one silica for the cpus to save the ghz being cranked up... what would we be at now, 8ghz + with 150c temps?

Well take Bulldozer as an example (Ye ye we all know how "crappy" that architecture is blah blah blah), they crancked 4GHz on stock on 32nm now just think how much would they bring on 28nm, its the strange thing that I don't get why Intel is not cranking up the GHz on their CPUs.

sagum said,

For sure, 22nm is already very close to the physical limits (16nm) before electrons are simply too big to work, regardless of what they do with it.

Still, if they are having problems going down to 22nm maybe they should be looking at alternative ways of getting more out of less. We already went the route of multiple die on one silica for the cpus to save the ghz being cranked up... what would we be at now, 8ghz + with 150c temps?

If 16nm is the limit, how is 11nm in the works?

aalchemy said,

electrons too big? uhm, what?... electrons are 5677906 times smaller than 16nm.

http://www.wolframalpha.com/in...m%2Fdiameter+of+an+electron

You are correct - it has nothing to do with electrons being too big. When you get down to things in the nanometer scale you're dealing with quantum mechanics which don't obey the normal physical laws and properties we're used to. The reason there's a limit is due to electron tunneling and sizes of the silicon gates.

Electron tunneling occurs when a material acts as a superhighway for electrons. They 'tunnel' right through it. What actually happens is the electron meets the silicon gate in the transistor on one side and appears on the other side of it as though the material weren't even there. In effect it teleports.

Since transistors are meant to control the flow of electrons this tunneling is an issue. You can't control the flow of something that doesn't even notice you're there. There's also the issue of conductivity of silicon at smaller sizes. We've actually reached silicons limit, we're just getting by because we 'dope' it with other elements. At 16nm the instances of electron tunneling will increase and the conductivity will go down. It happens (less commonly and in poor quality processors) at even 22nm, mostly the conductivity issue.

Breakthrough said,

I'm not doubting that they're having issues, but the reasons given in that article are complete bull****.

Can anyone explain to me how a chip drawing less power runs hotter?

The smaller microarchitecture is why it runs hotter (hence the TDP heading north). With smaller architectures, you have more transistors (look at the transistor count compared to either Sandy Bridge or Yorkfield, let alone Wolfdale, Kentsfield, or Conroe) in a smaller space. When you pack more transistors into a smaller space, heat dissipation becomes a larger issue. Just to maintain an identical systemwide TDP (Ivy vs. Sandy), Ivy will need more efficient cooling than Sandy. It's not a nightmare (even for Intel), as there are cooling systems that can deal with it that are not overly complex (closed-loop water-cooling, for example) except in terms of public relations - and that is exactly the problem for Intel.

KCRic said,
dealing with quantum mechanics which don't obey the normal physical laws and properties we're used to.

.... eeerr, yeah... (looks around) I was going to say that.
Thanks giving us the proper explanation

x-byte said,
Smaller process, more leakage on the chip. Which translates into heat.

I think everybody who responded forgot that every watt that goes into your computer eventually turns into heat. If you have two chips, with the same heatsinks, and one consumes 75W and the other 95W... It doesn't matter which has what design underneath, the one that consumes 95W will run hotter, since it's using more power (all of which turns into heat).

PGHammer said,
The smaller microarchitecture is why it runs hotter (hence the TDP heading north). With smaller architectures, you have more transistors (look at the transistor count compared to either Sandy Bridge or Yorkfield, let alone Wolfdale, Kentsfield, or Conroe) in a smaller space. When you pack more transistors into a smaller space, heat dissipation becomes a larger issue.

I agree, heat dissipation is a bigger issue - but we're talking about the same amount of heat. Those smaller transistors draw less power (since they literally have less capacity to do so), so the effect isn't as localized as you state. However, I agree, the heat dissipation of the entire chip is a problem.

One that is the same as every other processor in existence.

Also means you will have to check for driver availability for the chipset for your OS of choice when you do a CPU upgrade.

FoxieFoxie said,
Ivy Bridge is a failure so who knows what happens to this

Ivy Bridge hasn't even launched yet, could you atleast hold back your fanboy tendacies until it's available?!

BumbleBritches57 said,

Ivy Bridge hasn't even launched yet, could you atleast hold back your fanboy tendacies until it's available?!

Check link few posts below

Vice said,
This is good for everyone. Can't wait.

The real question is how much more will set us back per CPU may guest $300 more over Intel all ready is so over price it an't funny.

I, for one, am happy to see the N.Bridge + S.Bridge and PCH go away for good. It certainly complicates the picture when building your PC and the performance benefits of having all of the control in the CPU are noticeable. When I think back to the way that hardware used to be with n/s bridge, jumpers, crappy drivers, Windows 95/98/ME, config.sys, etc. it reminds me of how nice things are today.

Tim Dawg said,
I, for one, am happy to see the N.Bridge + S.Bridge and PCH go away for good. It certainly complicates the picture when building your PC and the performance benefits of having all of the control in the CPU are noticeable. When I think back to the way that hardware used to be with n/s bridge, jumpers, crappy drivers, Windows 95/98/ME, config.sys, etc. it reminds me of how nice things are today.

Hmm sound more like your just being Lazy rigth?.

Tim Dawg said,
I, for one, am happy to see the N.Bridge + S.Bridge and PCH go away for good. It certainly complicates the picture when building your PC and the performance benefits of having all of the control in the CPU are noticeable. When I think back to the way that hardware used to be with n/s bridge, jumpers, crappy drivers, Windows 95/98/ME, config.sys, etc. it reminds me of how nice things are today.

Sandy Bridge is from the 2000s not the 90s?

Shahrad said,
So does this mean there would be LFF, SFF and TFF (Tiny Form Factor) ??

Could very well be. But the nano/mini itx mothers boards are already seeing all in one chips for the remaining components to help reduce the size of the motherboards... so I'm wondering if the all in one cpu is hurting motherboard manufacturing attempts to produce anything other then the norm'.

We could very well see a very basic board with just IO ports and a CPU connected to it... its not hard to see a point where, due to the limits on silica miniaturization, we'll see additional extras included in the CPU to help speed things up. How about bumping up the L1 cache for example to a few GB rather then MB for example. Might seem crazy but its the kind of things we're going to see happening in the next few years as the way forward.

Adding more cores or phycial cpus to the boards won't help too much as most developers are struggling to make use of 4 cores efficently so something has to break the mould at some point.

sagum said,
Could very well be. But the nano/mini itx mothers boards are already seeing all in one chips for the remaining components to help reduce the size of the motherboards... so I'm wondering if the all in one cpu is hurting motherboard manufacturing attempts to produce anything other then the norm'.

Wrong there not complete integration not even the Pico-ITX
This bear min needs
1 Ethernet LAN
1 Video Output
1 Audio Output
1 or 2 USB Port depend on the Mouse, Touchpad or Pointer/Keyboard
2 SATA Port Harddrive/CDRom
1 Memory Slot

sagum said,
How about bumping up the L1 cache for example to a few GB rather then MB for example.

That becuse it is crazy on L1 that only slow it down and it in kb not mb even with L2 or L3 you can only add so much it most likey never get to GB.

sagum said,
... Adding more cores or phycial cpus to the boards won't help too much as most developers are struggling to make use of 4 cores efficently so something has to break the mould at some point

Only on non-MS OSes.
Microsoft are doing the heavy lifting for these developers of which you speak.
async and await == I don't care about cores, let the runtime make optimum use for me.