bit-tech.net

Intel pledges Skylake ramp in 2015

Intel pledges Skylake ramp in 2015

Intel has pledged to begin mass production of its 14nm Skylake family in the second half of 2015, despite the schedule slip suffered by predecessor Broadwell.

Intel has pledged to continue with plans to begin mass production of its next-generation Skylake chips in the second half of next year, despite the schedule slip that delayed predecessor Broadwell.

Broadwell, the successor to the current-generation Haswell microarchitecture, is based on a 14nm process node which has been giving Intel a spot of bother. Plans to begin mass production of Broadwell processors last year were postponed due to yield problems at the extremely small feature size required of the parts. Although since resolved, Broadwell is still hanging back with rumours claiming overstock of Haswell parts is staying Intel's hand.

The delays that have beset Broadwell may have a knock-on effect for its successor, Skylake. Detailed in a slide leaked in July last year, Skylake follows the process shrink of Broadwell with an updated microarchitecture at the same 14nm process node. Skylake will, the slide claimed, support PCI Express 4.0, Advanced Vector Extensions (AVX) 3.2 and DDR4 memory. Officially, Skylake has no formal launch date but those following Intel's earlier release schedules have expected a release some time in late 2015 to early 2016.

Although Intel refuses to comment on rumours surrounding its launch schedule, the company's chief executive Brian Krzanich has suggested that Skylake will be hitting the market within its originally-rumoured timeframe. 'We had a lot going on,' Krzanich claimed, in response to an analyst's query regarding Intel's use of Taiwan Semiconductor (TSMC) for SoFIA chip production, in his company's most recent earnings call. 'The ramp of Broadwell, the ramp of Skylake in the second half of next year, plus bringing these products inside.'

Krzanich also confirmed plans to transition its mobile parts, including the outsourced SoFIA heavily-integrated chip, to internal production on a 14nm process. These moves, Intel has claimed, will boost demand for its parts - but profitability for its loss-making Mobile and Communications Group is a long way distant. 'I'd say for 2015, I would expect to see reduction in the loss [of the group],' chief financial officer Stacy Smith added. 'Not profitability, but a reduction in the loss will feel pretty good when we get there and then we'll keep driving towards the long-term profitability goal.'

Sadly, Intel did not confirm any further details regarding Skylake - but if production ramp is planned for the second half of 2015, retail availability should not be far behind.

14 Comments

Discuss in the forums Reply
Spreadie 17th April 2014, 13:55 Quote
So, Haswell hasn't sold in expected numbers, Broadwell was late, but is still being held back because of high Haswell stocks, and they still plan to release Skylake on time?

Did they actually resolve the low yields on Broadwell or have they decided low stocks won't matter if they release Skylake on schedule?
Hustler 17th April 2014, 14:49 Quote
Here's an idea Intel,

Drop the price of the Haswell i5/i7 chips by $30 & $50 across the entire range today, and see that unsold stock clear in no time.

Of course....you wont do it though.
Guinevere 17th April 2014, 15:14 Quote
I really don't get the hate I've read on this forum against Intel for having the nerve to update their CPU range too often.

OMG, isn't that a really good thing? In the mobile space intel have taken us from VIA CPUs being a valid option for low power consumption chips to Intel being able to power a tablet 'all day' running a full fat windows install. This has happened in only a few short years. I had a UMPC running a VIA CPU.... it was a pig!

The Haswell range brought some real nice battery improvements on mobile. It was a worthwhile upgrade to make available. I didn't go out and replace my Ivy Bridge laptop due to the cost, but I would love the battery life update which was almost double the previous generation.

Would people be happier if they hadn't of bothered to update the desktop CPUs to Haswell and say moved to a two year update cycle?
Corky42 17th April 2014, 15:35 Quote
Quote:
Originally Posted by Guinevere
OMG, isn't that a really good thing? In the mobile space intel have taken us from VIA CPUs being a valid option for low power consumption chips to Intel being able to power a tablet 'all day' running a full fat windows install. This has happened in only a few short years. I had a UMPC running a VIA CPU.... it was a pig!

That depends if you are in the market for a low power CPU doesn't it ? Then again if you do want a low power draw CPU wouldn't you just choose ARM.
I'm not sure i would go as far as to say people hate Intel, but it seems to me (and maybe others) that Intel have neglected the desktop (enthusiasts) market for a good few years trying to chase the mobile (low power) market.
jrs77 17th April 2014, 15:41 Quote
Quote:
Originally Posted by Guinevere
I really don't get the hate I've read on this forum against Intel for having the nerve to update their CPU range too often.

OMG, isn't that a really good thing? In the mobile space intel have taken us from VIA CPUs being a valid option for low power consumption chips to Intel being able to power a tablet 'all day' running a full fat windows install. This has happened in only a few short years. I had a UMPC running a VIA CPU.... it was a pig!

The Haswell range brought some real nice battery improvements on mobile. It was a worthwhile upgrade to make available. I didn't go out and replace my Ivy Bridge laptop due to the cost, but I would love the battery life update which was almost double the previous generation.

Would people be happier if they hadn't of bothered to update the desktop CPUs to Haswell and say moved to a two year update cycle?

The refreshed Haswell-chips are unnecessary, as they offer nothing but a 100MHz increase. So yeah, some updates are really something we could do without.

Ofc I want to see improvement, but I wan't to see real improvements. The improvement between IvyBridge and Haswell are too small to call them a real improvement imho so I most of the time skip one generation and only buy the ticks instead of the following tocks.

From a buyers-perspective the small updates (tocks) makes the market overcrowded with unnecessary products, so yeah I'd prefer a situation with a two-year cycle, where there's only ticks and no tocks.

I'm in the market for a new setup, as I basically don't need a dedicated GPU anymore, but more CPU-power. However, I don't want to waste my money on a small upgrade and that's why I'm waiting for the new 14nm-parts to get shipped, as they noot only will have more grunt but they'll also use less power then the Haswell-parts.
Gareth Halfacree 17th April 2014, 15:48 Quote
Quote:
Originally Posted by jrs77
From a buyers-perspective the small updates (tocks) makes the market overcrowded with unnecessary products, so yeah I'd prefer a situation with a two-year cycle, where there's only ticks and no tocks.
Trouble is, the tick-tock thing is all about minimising the potential for screw-ups - not maximising the frequency of product launches (although I'm sure Intel hates that.) Doing a die shrink, as the Broadwell delay has proven, ain't easy; doing an entirely new microarchitecture with all-new features also ain't easy. Doing both at the same time: Eugh. Nightmare.

Tick-tock lets Intel work on creating a new microarchitecture, then making it work at a smaller process node, then use what it learnt to create a new microarchitecture on that same node before shrinking it down to an even-smaller node. Now, Intel could arguably still do this on a two-year cycle - spend a year building and proving the new architecture then another year shrinking it to a new process before releasing it - but that has significant risk involved. Getting it out of the door and in customers' hands at both stages of development, rather than just at the end of the development cycle, means Intel can see what works and what doesn't - and use that knowledge for the next version. 14nm proved a pain for Broadwell; the glitches should hopefully be ironed out in time for Skywell. Major errata in Sandy Bridge; fixed in Ivy Bridge.
schmidtbag 17th April 2014, 15:57 Quote
Quote:
Originally Posted by Hustler
Here's an idea Intel,

Drop the price of the Haswell i5/i7 chips by $30 & $50 across the entire range today, and see that unsold stock clear in no time.

Of course....you wont do it though.

For some of their products, it's going to be more than $50 off for me to be willing to pay. But to me the real problem with intel is it's getting increasingly difficult to justify an upgrade. Sandy Bridge is still more than sufficient for most purposes and still pretty power efficient.

I'd argue that intel should focus more on their IGPs, but I honestly don't think that really matters. When people buy an intel CPU, they're not buying it for the IGP. That being said, it'd bother me to know that the high premium I pay is partially going to something that a discrete GPU disables. Personally, I think intel should remove the IGP from their i7 products and improve it in everything else.
Corky42 17th April 2014, 16:52 Quote
Quote:
Originally Posted by jrs77
From a buyers-perspective the small updates (tocks) makes the market overcrowded with unnecessary products, so yeah I'd prefer a situation with a two-year cycle, where there's only ticks and no tocks.

Odd, i prefer the tocks.
As Gareth pointed out, normally the tock improves or fixes any problems discovered from the tick.
schmidtbag 17th April 2014, 17:04 Quote
Quote:
Originally Posted by Corky42
Odd, i prefer the tocks.
As Gareth pointed out, normally the tock improves or fixes any problems discovered from the tick.

Yeah, but maybe if intel had longer release cycles, such issues wouldn't occur in the first place. What exactly is intel in such a rush for anyway? With AMD so far behind the high-end CPUs, intel could use the same product for the next 3 or 4 years. At this point it feels like intel is just releasing new products simply because they can, but it doesn't seem their approach is very profitable lately considering the dropping sales.
jrs77 17th April 2014, 18:39 Quote
Quote:
Originally Posted by Gareth Halfacree
Trouble is, the tick-tock thing is all about minimising the potential for screw-ups - not maximising the frequency of product launches (although I'm sure Intel hates that.) Doing a die shrink, as the Broadwell delay has proven, ain't easy; doing an entirely new microarchitecture with all-new features also ain't easy. Doing both at the same time: Eugh. Nightmare.

Tick-tock lets Intel work on creating a new microarchitecture, then making it work at a smaller process node, then use what it learnt to create a new microarchitecture on that same node before shrinking it down to an even-smaller node. Now, Intel could arguably still do this on a two-year cycle - spend a year building and proving the new architecture then another year shrinking it to a new process before releasing it - but that has significant risk involved. Getting it out of the door and in customers' hands at both stages of development, rather than just at the end of the development cycle, means Intel can see what works and what doesn't - and use that knowledge for the next version. 14nm proved a pain for Broadwell; the glitches should hopefully be ironed out in time for Skywell. Major errata in Sandy Bridge; fixed in Ivy Bridge.

What major errata in Sandy Bridge?

And Broadwell will not be a die-shrink from Haswell, but a totally new architecture, as they remove the controller from the die, etc.

Also. They can take their time ironing out glitches in their offices as they like, but they don't need to release those tock-thingies to the public. A new CPU every two years is perfectly fine.
Gareth Halfacree 17th April 2014, 19:50 Quote
Quote:
Originally Posted by jrs77
What major errata in Sandy Bridge?
I was speaking hypothetically, although there are some eyebrow-raising errata (PDF) listed from page 22 onwards for the latest mobile chips. Then there was the bug in the Sandy Bridge supporting logic chipset - not the CPU itself, I'll grant you - which meant Intel had to recall eight million chips...
Quote:
Originally Posted by jrs77
And Broadwell will not be a die-shrink from Haswell, but a totally new architecture, as they remove the controller from the die, etc.
That doesn't alter the microarchitecture, just the design: Broadwell is the Haswell microarchitecture, with only minor changes. You won't get a new microarchitecture until Skylake. That's how tick-tock works.
Quote:
Originally Posted by jrs77
Also. They can take their time ironing out glitches in their offices as they like, but they don't need to release those tock-thingies to the public. A new CPU every two years is perfectly fine.
Too expensive; they need to build the chips anyway to prove the new microarchitecture/process node. If they built a dozen to test, they'd cost millions of dollars each; build a few million and sell 'em, and your test units cost a few dollars each. Which would you rather do? Then there's the fact that not all bugs will be found in the lab, but when you've got a few million units in the wild it's a different matter...
mi1ez 22nd April 2014, 00:50 Quote
A Skylake ramp? I can't be the only person who sees that phrase and thinks of Evel Knievel...
Yadda 22nd April 2014, 01:17 Quote
Quote:
Originally Posted by mi1ez
A Skylake ramp? I can't be the only person who sees that phrase and thinks of Evel Knievel...

:)

It reminded me of Alton Towers.
Nexxo 22nd April 2014, 08:57 Quote
Of course the advantage of each new release is that last year's almost as capable model drops a bit in price. And we like cheaper chips, don't we?
Log in

You are not logged in, please login with your forum account below. If you don't already have an account please register to start contributing.



Discuss in the forums