intel's newest 2013-2014 roadmap

[beginner99]

No Broadwell on Desktop makes sense and has nothing to do with AMD. Intel decided to fully go into mobile and it was well known already a year ago that Atom would move up from an older process node to the newest node.

So on 14 nm Atom (Airmont) will be first and then other mobile chips, eg. bga broadwell. Now for desktop power isn't much of an issue and iGPU performance doesn't matter much too due to easiness of installing a dGPU. So they can dedicate all 14nm capacity to mobile products an keep 22 nm factories running for desktop and maybe chipsets for 14 nm products.

makes perfect sense to me if intel would do this.

 

[mikk]

So on 14 nm Atom (Airmont) will be first and then other mobile chips, eg. bga broadwell. Now for desktop power isn't much of an issue and iGPU performance doesn't matter much too due to easiness of installing a dGPU. So they can dedicate all 14nm capacity to mobile products an keep 22 nm factories running for desktop and maybe chipsets for 14 nm products.

This first 14nm products will be Broadwell U-/Y.

 

[AtenRa]

22nm Atoms will be launching late 2013(Q4) early 2014(Q1), no way they will launch 14nm Atoms in H1(Q2) 2014. I could see a late 2014(Q4) if they will puss it.

 

[PhlashFoto]

I would guess we will see Broadwell sometime in the fall of 2014. It gives OEMs time to sell off inventory during back to school season. Plus I could see Intel wanting to have something to show and talk about at IDF in September.

I think any "Haswell refresh" would be model updates to current SKUs and not a platform refresh to delay 14nm for another year.

 

[MisterMac]

Intel could have delivered consistently higher IPC by putting the massive eDRAM cache on every Haswell, not just GT3e. But they didn't because they like segmentation and margins.

Ding.

They have plenty of innovations (probably some of the up sleeve) in terms of the overall absolute performance race - where they could have raised the bar each uARCH generation.

But they choose not to - and they're now dancing on the absolute thinnest incremental increases they can get away with.
Or atleast it seems this way.


You can easily argue im a cheapskate - because i don't feel LGA 2011 is worth it.

You could also argue - that by not giving excess multi-parallel\raw ipc power availeble to the general public neither will anyone pioneer or use power for any sector of computing to advance the general bar.

I'm worried absolute performance will stagnate now.

 

[Vesku]

I seem to recall some giant thread last year where a poster was adamant that the lack of competition from AMD would not hurt consumers at all. That even prior to Bulldozer, Intel was more competition to itself so everything would be A OK. Reason I remember is the thread sort of spiraled out of control after that, hehe.

Haven't heard that sales pitch in some time...

 

[MisterMac]

I seem to recall some giant thread last year where a poster was adamant that the lack of competition from AMD would not hurt consumers at all. That even prior to Bulldozer, Intel was more competition to itself so everything would be A OK. Reason I remember is the thread sort of spiraled out of control after that, hehe.

Haven't heard that sales pitch in some time...

Depends how you look at the picture.

From our enthuast absolute perf point - well...it's not worse (See your reference to BD for that ).

From perf\watt mobile - WOOOOOOOOHOOOOOOOO.

The haswell mobile gains are impressive, very impressive.
Again haswell is structured to the market in 2013 - and that is a more mobile world.


I geuss some of us still expected alot of the new ports and branch prediction logic to really kick even legacy ass.

 

[meloz]

I find it hard to believe we will not see small-socket Broadwell sometime in 2014. It might be pushed later into the year, but they will very likely launch it in 2014.

What is more likely is that the release date has not been firmed. They cannot settle on a release date because -for the first time- Atom (Airmont) will get the first use of newest manufacturing process (14nm).

Keep in mind Intel supply CPUs to >90% of the desktop and notebook market. They need to be able to produce a minimum amount to satisfy OEM demand. If Intel wins lots of high volume customers for Airmont they might not have the capacity to manufacture Broadwell alongside in high numbers.

So they might decided to first satiate the Airmont demand, then launch Broadwell. And since no one knows how well or poorly Airmont will do -it will all come down to pricing- no one can be sure when Broadwell will be launched.

I think any "Haswell refresh" would be model updates to current SKUs and not a platform refresh to delay 14nm for another year.

I have the same feeling.

 

[A5]

Ding.

They have plenty of innovations (probably some of the up sleeve) in terms of the overall absolute performance race - where they could have raised the bar each uARCH generation.

But they choose not to - and they're now dancing on the absolute thinnest incremental increases they can get away with.
Or atleast it seems this way.


You can easily argue im a cheapskate - because i don't feel LGA 2011 is worth it.

You could also argue - that by not giving excess multi-parallel\raw ipc power availeble to the general public neither will anyone pioneer or use power for any sector of computing to advance the general bar.

I'm worried absolute performance will stagnate now.

They're not pushing max IPC all the time because they have to keep power consumption in check for notebooks.

People here forget that Apple probably buys more Intel chips in a quarter than the entire "enthusiast" market does over the whole year. They're designing their chips based on what their real customers are asking for, which is lower power and a better iGPU.

 

[Phynaz]

I seem to recall some giant thread last year where a poster was adamant that the lack of competition from AMD would not hurt consumers at all. That even prior to Bulldozer, Intel was more competition to itself so everything would be A OK. Reason I remember is the thread sort of spiraled out of control after that, hehe.

Haven't heard that sales pitch in some time...

We've known for some time that Haswell is meant to compete in the mobile space. I still don't get why people don't understand this.

 

[R0H1T]

I seem to recall some giant thread last year where a poster was adamant that the lack of competition from AMD would not hurt consumers at all. That even prior to Bulldozer, Intel was more competition to itself so everything would be A OK. Reason I remember is the thread sort of spiraled out of control after that, hehe.

Haven't heard that sales pitch in some time...

Maybe this is the thread you're talking about, except I was arguing that with AMD gone the x86 market would shrink faster & most others said that it'd not be the case :\
TSMC and ARM Tape-Out First ARM Cortex-A57 Processor on 16 nm FinFET Technology

Keep in mind Intel supply CPUs to >90% of the desktop and notebook market. They need to be able to produce a minimum amount to satisfy OEM demand. If Intel wins lots of high volume customers for Airmont they might not have the capacity to manufacture Broadwell alongside in high numbers.

Definitely not true

They're not pushing max IPC all the time because they have to keep power consumption in check for notebooks.

Again its not because they wish to but the case being there's no real gain in improvising the age old core microarch, people will really have to accept that Intel can only do so much, as against the relatively new Bulldozer microarch which still has quite a bit of juice left in it to extract. Silvermont will be interesting though because Intel claims its a redesign so I'm expecting some radical changes & not just i3/i5 slapped onto a smaller die !

You think such a move wouldn't help them in the server market competing against the likes of IBM, HP, ORACLE et al ?

 

[A5]

Again its not because they wish to but the case being there's no real gain in improvising the age old core microarch, people will really have to accept that Intel can only do so much, as against the relatively new Bulldozer microarch which still has quite a bit of juice left in it to extract. Silvermont will be interesting though because Intel claims its a redesign so I'm expecting some radical changes & not just i3/i5 slapped onto a smaller die !

Don't get your hopes up.

Bulldozer isn't really that different from a uarch perspective, it just has 2 ALUs for each FPU.

Incremental changes to widen the pipeline everywhere (like AVX2) are what you're going to see going forward. Intel has no reason to do a clean-sheet design.

You think such a move wouldn't help them in the server market competing against the likes of IBM, HP, ORACLE et al ?

They have other products to cover that market. Haswell Xeons won't be competing against anything those companies are selling. Hell, the high-end Xeons (everything above the E3 series) are still based on SB.

 

[Fjodor2001]

Two possible reasons for Intel changing their mind and deciding to not produce any 14 nm Broadwell desktop CPUs:

* They have problems with the 14 nm process (e.g. heat issues due to more transistors per die area), so they cannot push the CPUs up to high enough frequencies. We've already seen Intel starting to hit a frequency wall with SB and Haswell. Using the process for lower frequency Atom and ULV parts works ok though.

* They consider the low power / mobile segment more important. And to be able to compete with ARM they dedicate the 14 nm process to those CPUs.

 

[wlee15]

Don't get your hopes up.

Bulldozer isn't really that different from a uarch perspective, it just has 2 ALUs for each FPU.

Incremental changes to widen the pipeline everywhere (like AVX2) are what you're going to see going forward. Intel has no reason to do a clean-sheet design.



They have other products to cover that market. Haswell Xeons won't be competing against anything those companies are selling. Hell, the high-end Xeons (everything above the E3 series) are still based on SB.

There's a dieshot of an AMD module which appear to have dual 256-bit FMA and upgrade the integer units to have 4 ALU or AGU rather than the existing 2 ALU and AGU configuration. The 256-bit part seems to conflict with last years Steamroller Hotchips presentation, however the improvement to the integer do not and they seem reasonable considering the addition of dual decoders in Steamroller.

 

[AtenRa]

Just obtained the following slides from Intel,

IV-E socket 2011 coming end of Q3, Core i3 Haswell coming start of Q3

njoy

 

[NTMBK]

Just obtained the following slides from Intel,

IV-E socket 2011 coming end of Q3, Core i3 Haswell coming start of Q3

njoy

Interesting, looks like there will still be Haswell Celerons and Pentiums, even with the new Silvermont based ones. That's not going to get confusing at all :thumbsup:

 

[JTsyo]

Wiki

Seems Skylake will be the CPU that's out when I'm looking to upgrade.

 

[krumme]

Just obtained the following slides from Intel,

IV-E socket 2011 coming end of Q3, Core i3 Haswell coming start of Q3

njoy

Thanx

Baytrail feb to april 2014 ??

 

[lagokc]

Two possible reasons for Intel changing their mind and deciding to not produce any 14 nm Broadwell desktop CPUs:

* They have problems with the 14 nm process (e.g. heat issues due to more transistors per die area), so they cannot push the CPUs up to high enough frequencies. We've already seen Intel starting to hit a frequency wall with SB and Haswell. Using the process for lower frequency Atom and ULV parts works ok though.

* They consider the low power / mobile segment more important. And to be able to compete with ARM they dedicate the 14 nm process to those CPUs.

I'll give you a third:

*Broadwell reduces power consumption while offering a 0% performance improvement over Haswell. Since this doesn't matter for desktops, Intel is taking the opportunity to sell off their remaining stock of Haswell instead of creating superfluous part numbers.

 

[mrmt]

*Broadwell reduces power consumption while offering a 0% performance improvement over Haswell. Since this doesn't matter for desktops, Intel is taking the opportunity to sell off their remaining stock of Haswell instead of creating superfluous part numbers.

Regardless of performance they have die size. For Intel it is worth to make a die shrink because of the smaller die for the desktop parts. If they are not going to ramp up desktops with 14nm, this either means:

- They plan a slower 14nm ramp up

- They have other plans for 14nm capacity (Atom probably)

 

[krumme]

Regardless of performance they have die size. For Intel it is worth to make a die shrink because of the smaller die for the desktop parts. If they are not going to ramp up desktops with 14nm, this either means:

- They plan a slower 14nm ramp up

- They have other plans for 14nm capacity (Atom probably)

It makes good sense to do both of the above imho.

What Intel have because of safe profitablity on the server market, is loads of maneuverbility and control at least on the 3-5 year perspective.

Intel only needs 14nm one place and thats Atom.
Every other market they just own hands down.
They absolutely have to put the pedal to the metal for Atom and go all the way for their mobile strategy to succeed.
Unlike AMDs jaguar, they havnt loaded atom with any heavy fpu or features, so it will perhaps scale better with denser process?

Secondly, if their mobile attack does not execucute quite as expected, already making a slower ramp of 14nm will give them the medium future cash to maintain control. They can not give Atom 22nm and 14nm away for free for several years as they subsidice Atom now. It just looks like the last boy at the dance floor.

Intel does not only have the best process tech as it is, but they have world class competences creating it. Its not only heads, its the 1+1=3 effects in having it for a long time. Thats the true value of the company and i dont think we see it reflected in shareprice.
If they do learn to understand they can more than just beeing a fab on the long term, perhaps they can steer the company in a direction, and create products, with even more profitability.

The combination of financial maneuverbility and worldclass competences sealed in culture, it Intels best competitive advantage.

 

[mrmt]

Intel only needs 14nm one place and thats Atom. Every other market they just own hands down.

They absolutely have to put the pedal to the metal for Atom and go all the way for their mobile strategy to succeed. Unlike AMDs jaguar, they havnt loaded atom with any heavy fpu or features, so it will perhaps scale better with denser process?

SoC size for Bay trail is supposed to be rather small, so I think they might have a winner there. 14nm might be the key for an uncontestable advantage in die size, on top of the performance advantage they have with Atom.

The Atom hypothesis also fits 14nm schedule. They are going to ramp up 14nm on Q413 with Broadwell. The second line up should be Airmont, and with that there is no capacity for a desktop ramp up.

 

[NTMBK]

Regardless of performance they have die size. For Intel it is worth to make a die shrink because of the smaller die for the desktop parts. If they are not going to ramp up desktops with 14nm, this either means:

Intel are already struggling to fill their fabs- smaller dies isn't necessarily a big plus for them. But keeping their 22nm fabs churning out high margin parts for an extra year? Clear win.

 

[mrmt]

Intel are already struggling to fill their fabs- smaller dies isn't necessarily a big plus for them. But keeping their 22nm fabs churning out high margin parts for an extra year? Clear win.

More thoughts about the subject:

- Chipset manufacturing will soon be gone, because only server parts will use it, so 32nm chipset demand will be far smaller than in previous nodes, meaning that 32nm will have a quick phase out and 22nm is likely to have a far quicker phase out because there will be no significant chipset market to talk about.

- Atom is moving up from lagging edge nodes to the bleeding edge node, further eroding the demand in the lagging edge node.

- In order to compensate for the chipset and Atom volume losses in the lagging edge node, it might make sense for Intel to move the desktop line up from the bleeding edge node.

If those assumptions are correct, Intel might not be skipping Broadwell in the desktop line up, but just delaying it. We'll have mobile Skylake and desktop Broadwell at the same time.

 

[krumme]

More thoughts about the subject:

- Chipset manufacturing will soon be gone, because only server parts will use it, so 32nm chipset demand will be far smaller than in previous nodes, meaning that 32nm will have a quick phase out and 22nm is likely to have a far quicker phase out because there will be no significant chipset market to talk about.

- Atom is moving up from lagging edge nodes to the bleeding edge node, further eroding the demand in the lagging edge node.

- In order to compensate for the chipset and Atom volume losses in the lagging edge node, it might make sense for Intel to move the desktop line up from the bleeding edge node.

If those assumptions are correct, Intel might not be skipping Broadwell in the desktop line up, but just delaying it. We'll have mobile Skylake and desktop Broadwell at the same time.

During the night my S4 loses about 3% percent of its battery life - so improvements from Atom Q2 2014 is hardly noticiable at all. If its even better than the non ooo that is running at that time.

The speed on the S4 is imho not noticiable different from the S3 or S2 for that matter i had before. I wonder how they manage to convince people it does. But Atom hardly makes a difference here - quite the opposite it have fewer cores (think tegra 3 sales success).

For battery life what matter is 75% screen efficiency, as is easily seen from the powerusage.

For performance what matters is mostly software. And when hardware performance matters its on the gpu part because whats driving the market here is gaming. My kids dont play any games that does not run perfectly on a TI 4470 as it is (dual core A9 with relatively slow gpu).

Even when asuming Apple and Samsung didnt exist in this world, - Intels ambition to sell mobile solution here seems just short of impossible. Then add who the customers is, and it becomes completely incomprehendable for me. When Samsung buys from them its just to put pressure on Qualcomm or because they are short one some cheap stuff for their cheap tablets. They are using them. I simply dont understand it.

They are entering a market that is soon going into maturity for the cpu side, and the customers really dont like to buy cpu from them.

I think Intel needs to do more radical changes to control its future capex. Even if they did enter and get into fx. 10% of the mobile shipments, the story would be the same. Small dies, small user benefit. Less money for proces node development.

Its like they are looking everywhere to find a business to apply their foundry capacity and technology to stubborn to imagine its not there.