Discussion Intel current and future Lakes & Rapids thread

[Yotsugi]

Lol Yotsugi - your coping mechanism is funny

You're really bad at deflecting too.

 

[liahos1]

Who is deflecting? Look in the mirror.

 

[Yotsugi]

Who is deflecting? Look in the mirror.

You are.
Being fair, you being funny at forums is better than Intel being unfunny with their damage control pieces but alas.

 

[Gideon]

IMO the biggest fail @ Intel isn't even the ongoing saga of 10nm nor the difficulty of backporting Sunny Cove. It's their total inability to produce even a slightly improved CPU arch on 14nm after all these years (6-7!). ARM releases a new core every year (often with minimal node improvements) Nvidia managed to double their perf/watt with maxwell, on the same node, etc ...

 

[JoeRambo]

IMO the biggest fail @ Intel isn't even the ongoing saga of 10nm nor the difficulty of backporting Sunny Cove. It's their total inability to produce even a slightly improved CPU arch on 14nm after all these years (6-7!). ARM releases a new core every year (often with minimal node improvements) Nvidia managed to double their perf/watt with maxwell, on the same node, etc ...

Indeed, Intel could counter AMD on 14nm with some sort of improved core built without GPU and taking ~275mm^2. Nehalem was 263mm^2. Fit 16 of them in desktop CPU with "130 Intel watt" socket and call it a day. Heck add extra pain for AMD by pricing it $266 and doing triple channels of mem for extra salt content.

 

[NostaSeronx]

For all your information, Sunnycove is a 14nm original. It was forward ported to 10nm, it doesn't need to be backported.

Skylake -> Sunnycove(14nm) -> Willowcove(10nm)
vs
Skylake -> Cannonlake -> Icelake

However, the current roadmap has Willowcove not being the first 10nm(ignoring failures), but the last 10nm architecture.

Not sure if anyone cares: Sapphire Rapids/Granite Rapids are only planned to be AZ fabbed initially.

 

[Cardyak]

IMO the biggest fail @ Intel isn't even the ongoing saga of 10nm nor the difficulty of backporting Sunny Cove. It's their total inability to produce even a slightly improved CPU arch on 14nm after all these years (6-7!). ARM releases a new core every year (often with minimal node improvements) Nvidia managed to double their perf/watt with maxwell, on the same node, etc ...

This is what leads me to believe that future microarchitectures from the Cove ranges will have substantial increases in IPC, Intel has an entire legion of engineers - they can’t just be sitting around doing nothing.

There’s been a dearth of architecture innovation from Intel since Skylake, but I expect Sunny Cove is just the start of a quick succession of new designs which are effectively sitting in the backlog. The last few years have been a drought in regards to IPC improvement, but looking at Intel’s future roadmap there may be a flood coming over the next few years with a yearly cadence of Cove updates.

 

[Dayman1225]

Wow we totally should trust Intel, for they never ever lied to us right

Don’t shoot the messenger

 

[Yotsugi]

Don’t shoot the messenger

Yeah, where I live we prefer stabbing people.



Inappropriate post for the tech forum.


esquared
Anandtech Forum Director

 

[.vodka]

Yotsugi, just because you and Charlie says something is so, doesn’t make it so.

Charlie has been reporting on how inviable 10nm was, is and will be, for years now. What he reported is what ultimately happened. Cannonlake was not manufacturable, Icelake not much better in that regard. At least there's some volume for some notebooks.

At this point, Intel spinning the leak as if it's false, makes it true. Whatever comes out of those 10nm (or whatever they're calling 10nm now, as the original version is dead and buried) fabs won't be available on desktop.

I guess they're also getting their chiplet strategy in shape because I don't see this dumpster fire of a node yielding those big core count dies Intel loves to use in their Xeons.

At least 7nm has EUV available for it, to make things a bit easier. It shouldn't miss the mark by much.

 

[Ajay]

You're really bad at deflecting too.

Who is deflecting? Look in the mirror.

What the heck, are we in high school? Please stop.

 

[Yotsugi]

What the heck, are we in high school? Please stop.

Hey, it's fun.

 

[Ajay]

I guess they're also getting their chiplet strategy in shape because I don't see this dumpster fire of a node yielding those big core count dies Intel loves to use in their Xeons.

Last I read, Ice Lake -SP was going to be a 26 core design with 22 enabled cores and very poor yields. Intel finally decided to ramp up both 10nm fabs, so they've got something going - mostly mobile I would expect. I'd be surprised if SR was bridged cores instead of EMIB, after all the work (and bragging) Intel has put into this - that would just be a pathetic failure of management. Speaking of Management, I think Intel has been going down hill, since Andy Grove left. Brian Krzanich just happened to be the anchor that sunk the boat with all his crony appointments.

I don't think Keller has been at Intel long enough to have much affect on Granite Rapids - especially having to deal with Intel's huge teams and bureaucracy. Anyway, 7nm is make of break for Intel - I'm sure that they have thrown everything including the kitchen sink into it and their CPU architecture. It will be sad if they stumble again.

 

[CHADBOGA]

is this all because Brian Krzanich happened?

He can't be the only incompetent, too much as gone wrong for it just to be down to Krzanich alone.

 

[mikk]

I was going to say desktop versions of any Atom parts that get released.

Even in this case it is in conflict with hardwareluxx. They explicitly told all desktop CPUs manufactured in 10nm are affected. Realistically the hardwareluxx news is a hoax and we will get Alder Lake-S as planned.

 

[NostaSeronx]

There is Tigerlake-Y replacing Atom in IoT/Automotive groups, Tigerlake-U replacing Icelake-U in PCCG, then there is Tigerlake-S sharing its launch with Cometlake-S. There is even potentially a Tigerlake-S refresh w/ USB4/PCIe4 that would launch with Alderlake-S.

¯\_(ツ)_/¯
Tigerlake(semi-custom) is Intel's golden SoC at Apple.

 

[jpiniero]

Even in this case it is in conflict with hardwareluxx. They explicitly told all desktop CPUs manufactured in 10nm are affected. Realistically the hardwareluxx news is a hoax and we will get Alder Lake-S as planned.

To hardwareluxx, the S line is the desktop. Which technically isn't true, as I mentioned since Intel sells 10W desktop versions of Atom. I thought Elkhart Lake was for IoT but it does have an IGP onboard.

 

[Zucker2k]

IMO the biggest fail @ Intel isn't even the ongoing saga of 10nm nor the difficulty of backporting Sunny Cove. It's their total inability to produce even a slightly improved CPU arch on 14nm after all these years (6-7!). ARM releases a new core every year (often with minimal node improvements) Nvidia managed to double their perf/watt with maxwell, on the same node, etc ...

It probably speaks to the progress they've been slowly and painfully chalking with 10nm to still want to stick with the node rather than backport sunnycove to 14nm. On the other hand, Intel is all about making money, so if they can get away with 14nm for another generation and still end up making the most money out of the current situation, then that's the path they'll choose.

 

[jpiniero]

It probably speaks to the progress they've been slowly and painfully chalking with 10nm to still want to stick with the node rather than backport sunnycove to 14nm.

Backporting a core would require a considerable amount of money and resources, not to mention time. Of course selling Skylake in 2021 is really not going to go over well.

 

[beginner99]

He can't be the only incompetent, too much as gone wrong for it just to be down to Krzanich alone.

But he is a the top and if the top guy decides to hire incompetent managers and is more a bean counter than tech person, you can imagne were it goes. Not paying the good engineers enough money for one. In in this advanced tech one brilliant mind can make all the difference.

 

[Gideon]

Backporting a core would require a considerable amount of money and resources, not to mention time. Of course selling Skylake in 2021 is really not going to go over well.

I'm well aware that it's really easy to be an armchair general in hindsight, but still ....

Considering the amount of resources Intel did have, it would have made sense to have 2 backup plans for 10nm in the beginning of 2017:
1. Coffee-Lake and derivatives (as was done)
2. A new arch on 14nm (less-aggressive than Sunny cove ,but using parts of it) for 2019-20 just in case 10nm is delayed even more.

They had enough design teams and money to pull that off. They already knew, at the very least in 2018 (possibly earlier) that even when 10nm chips finally start rolling out, the yields will be bad enough that they'll also need complementary 14nm products. That's why we have Comet-Lake, Cooper-lake, etc on top of Ice Lake.

Now, if everything went super well (which it didn't), it would have been a wasted effort. But even if it would have been only a bit better than the reality (say Ice-Lake would have come a quarter sooner with a bit better yields), it would still have been worth it.

They would at least have more competitive desktop and server parts with better mind-share (more IPC), rather than just bumping clock-speed way above the comfort zone on the Voltage/Frequency curve.

 

[maddie]

Backporting a core would require a considerable amount of money and resources, not to mention time. Of course selling Skylake in 2021 is really not going to go over well.

Anyone ever did a hike where the locals kept saying the destination was right around the next corner?

 

[DrMrLordX]

Intel denies the rumours and says

Intel: Yes, There Will be 10nm Desktop CPUs (Updated) | Tom's Hardware

A rumor emerged on Monday that Intel might have no plans at all for desktop processors on 10nm, and would bring 7nm to the desktop in 2022. However, Intel refuted the rumors and says its 10nm roadmap still includes desktop products.

www.tomshardware.com

I just wanted to add that here is a non-amp link:

Intel: Yes, There Will be 10nm Desktop CPUs (Updated)

Intel refutes rumors that it is skipping the 10nm process for desktop processors

www.tomshardware.com

(Waterfox choked on that amp link)

Also, note the clarification from Tom's posted as of 8 pm PDT last night:

Update 10/14/2019 8:00pt: Intel confirmed to Tom's Hardware that its use of the "Desktop Products" in its statement means "Desktop CPUs," specifically. It's noteworthy that the statement could equate to NUC processors only, and numerous other possibilities abound (like 10nm CPUs for the HEDT market only), but Intel hasn't provided further clarity.

So say hello to your Tiger Lake NUCs folks!

Last I read, Ice Lake -SP was going to be a 26 core design with 22 enabled cores and very poor yields.

That was the . . . I won't say original plan, that was the plan announced maybe late last year? Then earlier this year they came out and announced 36c chips too (thanks for the reminder @Yotsugi , I got the core count wrong) but they pushed ICL-SP out again at the same time. And then they tried to distract everyone with Cooper Lake to make it all sound lovely.

Intel finally decided to ramp up both 10nm fabs, so they've got something going - mostly mobile I would expect.

It looks like Intel is going to go full-bore with IoT and comm-level gear based on 10nm tech wherever possible. Lots of little Atom variants. Snow Ridge, Tanner Ridge, Elkhart Lake, Lakefield (okay, that isn't comm or IoT), and then enigmatically, there's . . . IceLake-D.

 

[jpiniero]

It looks like Intel is going to go full-bore with IoT and comm-level gear based on 10nm tech wherever possible. Lots of little Atom variants. Snow Ridge, Tanner Ridge, Elkhart Lake, Lakefield (okay, that isn't comm or IoT), and then enigmatically, there's . . . IceLake-D.

The Atoms might work, if they keep it small enough. 4 Tremont cores and 32 EUs would be a pretty small die, and they could cut to 2 cores and 8 EUs.

 

[mikk]

NUC devices are using mobile ULV CPUs, and by design this is mobile. I haven't seen a desktop roadmap with a NUC.