Intel "Rocket Lake" an Adaptation of "Willow Cove" CPU Cores on 14nm?

[btarunr]

The "Willow Cove" CPU core design succeeds "Sunny Cove," Intel's first truly new CPU core design in close to 5 years. "Sunny Cove" is implemented in the 10 nm "Ice Lake" microarchitecture, and "Willow Cove" cores are expected to debut with the 10 nm+ "Tiger Lake." It turns out that Intel is working to adapt "Willow Cove" CPU cores onto a 14 nm microarchitecture, and "Rocket Lake" could be it.

Twitter user @chiakokhua, a retired VLSI engineer with high hit-rate on CPU microarchitecture news, made sense of technical documents to point out that "Rocket Lake" is essentially a 14 nm adaptation of "Tiger Lake," but with the iGPU shrunk significantly, to make room for the larger CPU cores. The Gen12 iGPU on "Rocket Lake-S" will feature just 32 execution units (EUs), whilst on "Tiger Lake," it has three times the muscle, with 96 EUs. "Rocket Lake" also replaces "Tiger Lake's" FIVR (fully-integrated voltage regulation) with a conventional SVID VRM architecture.



We know from an older report that the 14 nm "Rocket Lake-S" silicon has up to 8 CPU cores, even as its predecessor, "Comet Lake-S," dials core-counts up to 10. We know now that the lowered core count is a trade-off for big IPC gains. For the desktop platform, "Rocket Lake-S" could herald the first major IPC uplift on the Intel platform since "Skylake."

View at TechPowerUp Main Site

 

[dj-electric]

If this is true - this is really disastrous, and all of us should worry about this. Progression from both sides might slow down because of this.

Willow Cove on 14nm... this is not good...

 

[Uroshi]

Maybe it is good. 14nm has proven to allow higher frequencies in power "unconstrained" situations as of now. Ryzen could be an exception (and maybe could be clocked higher at Intel 14nm equivalent technology).

Who said that the constant shrinking of the transistors will also mean higher frequencies? Maybe improving IPC on a "tweaked" 14nm process could be better for pure single core performance in the future.

I say, good job, to Intel for using their money wisely and cover both options, new 14nm CPUs and new processes 10+++nm / 7nm.

Edit: Replaced IPC where I used it wrong ...

 

[Tomorrow]

Makes sense to me. I've said before that Intel has new arch on 10nm but the process holds them back, so the IPC gain is negated by the lower clocks and core counts. On the other hand 14nm is mature and high clocking but stuck on old arch.

So they either need to fix 10nm or backport the new arch to 14nm.
Since they are backporting Willow Cove to 14nm this sends a clear message that 10nm is unfixable on desktop and will continue to be used only in laptops.
This despite what they recently said about rumors of 10nm not coming to desktop. Remember - Intel considers their NUC's as desktop. So putting a 10nm chip in a NUC they can say we did release 10nm on desktop.

Sometimes it is telling what is not being done as opposed to what is being done when it comes to future plans.

 

[john_]

It's because 10nm are ramping up and we will soon have a plethora of 10nm Intel CPUs for desktop PCs. Right?

Anyway, Intel will try to keep the crown for gaming and programs that don't see more than 8 cores/ 16 threads. At least they have a plan. They leave the market that needs more cores to AMD and try to keep what they can try to keep.

 

[_Flare]

Every 10nm (+++) product will be very very low volume. 10nm is broken and will remain so, until succeeded by 7nm whatsoever.
They only need to produce any 10nm stuff to not get sued.
It is a very logical step to minimize cost by "backporting" Willow Cove and have useable volume products, until any new process is ready.
They go a similar way like AMD by making the caches bigger and rise their associativity.

 

[theGryphon]

Some (most?) of you guys have a totally wrong understanding of what IPC is. IPC= instructions per clock. Higher clock frequency does zilch for IPC.

Call it single - core performance, call it raw performance, just, for the love of God, don't call it IPC!

 

[Midland Dog]

good thing that rocket lake wont have fivr, that means we get an apples to apples in clocks to see if backporting a uArch makes worse clocks

 

[fancucker]

14nm WC would produce a significant uplift and remember that backporting is extremely expensive so its definitely a worthwhile decision from a performance perspective

IPC wise at the most suboptimal rates according to SPECint/FP would put it above Zen 2. It'll be nice watching that 8core STILL beating AMD's ryzen 4XXX in ST applications and the core obsession take a halt to true single-core performance

 

[Deleted member 158293]

Intel declared profits instead of updating their R&D and now need to catch up. They're behind and will be the budget option for a while. That's how markets should work, at least nominally in a duopoly situation like this.

 

[Grog6]

Does it have fixes for released flaws?

 

[Vayra86]

Woopsie.

I'll say it again. 10nm performance chips are not happening

 

[yeeeeman]

14nm WC would produce a significant uplift and remember that backporting is extremely expensive so its definitely a worthwhile decision from a performance perspective

IPC wise at the most suboptimal rates according to SPECint/FP would put it above Zen 2. It'll be nice watching that 8core STILL beating AMD's ryzen 4XXX in ST applications and the core obsession take a halt to true single-core performance

Sunny Cove is already a few percent better than Zen 2. So Willow Cove should easily be a few percents better than Zen 3.

Does it have fixes for released flaws?

I really don't care about that..

 

[bug]

There's only so many ways I can say "yaaaawn".

 

[notb]

To all "oh no, another 14nm".
The other company sells 5 times less chips - big part of which are still previous generation 14nm models.
And they still fight to save the precious 7nm by making I/O dies on older node.

Intel will keep using 14nm for many years to come. If you don't understand it, at least stop writing these boring posts under every "Intel 14nm news".

A typical example - from a honorable gentlemen and a senior member:

There's only so many ways I can say "yaaaawn".

Yes, I'm sure. But what's the point?
They're doing another generation. Slightly better performance. Some fixes. Some new features. What's wrong with that?

It doesn't tickle your "enthusiastic needs" - I understand that. But processors are running our civilization. We need a lot of them. Some will not be "fast" nor "interesting". They just have to do their job.

 

[bug]

Yes, I'm sure. But what's the point?
They're doing another generation. Slightly better performance. Some fixes. Some new features. What's wrong with that?

It doesn't tickle your "enthusiastic needs" - I understand that. But processors are running our civilization. We need a lot of them. Some will not be "fast" nor "interesting". They just have to do their job.

The point is, they have real work to do
Would you dare someone without a PhD in CS to pick between an i5-1035G7 and an i7-10510U?

 

[Deathy]

Sunny Cove is already a few percent better than Zen 2. So Willow Cove should easily be a few percents better than Zen 3.

Cool, reviews are out then, right? Got some links for me of actual products being tested and compared like you talked about?

 

[R0H1T]

Sunny Cove is already a few percent better than Zen 2. So Willow Cove should easily be a few percents better than Zen 3.

And what logical property is that

 

[_Flare]

In the latest interview of anandtech with raja koduri, he said that the 7nm process used for ponte vecchio and gpus in general has somewhat lower requirements than for cpus.
As i said earlier, some very few "specialized" 10nm parts will happen, but very trimmed, pruned how you may call it, so the announced and still announced parts get shipped, to get not sued.
I could bet that intel has started to change some 10nm DUV facilities to 7nm EUV.
It sounded like intel will build ponte vecchio in 7nm completely on its own, but i´m not sure.
If they go all-in with gpus they need more building capacity, because the gpus and cpus won´t use the same process, thats obvious now.
Struggling with their 14nm supply tells that actually a big margin of potencially active facilities do nothing, but costing money by multiple time beeing rebuild for new (failed 10nm) processes.
Eventually the 10nm products will be available only a few days from one single facility-run and then the facility gets converted to some 7nm process, likely EUV.

 

[kakandung]

14nm++++ Lakes are way too refined and it's becoming too good (for desktop TDP), kinda like Windows XP was.
10nm desktop SKU simply won't live up to the expectation, if it compared to the well polished 14nm++++++++ (like Vista was to XP)

 

[Vya Domus]

Let's see, what are their options : Use the newer node but face potential performance regressions due to lower clocks and worse yields, or, use the older node, probably could squeeze more performance and better yields out of it but at the same time power consumption becomes rampant.

It's not great. I never looked forward with much excitement for anything Intel related in the last few years and this only deepens my discontent with them.

 

[ppn]

Sunny cove is 0-40% better than skylake depending on the task. 20% on average. Willow cove is up to 30% compared to sunny. So 15% on average. This means 1,2*1,15 on average and up to 1,4*1,3 compared to sky. This is like sandy bridge against pentium 4. I don't care if it is 14nm since the density of AMD chips is 50 M transistors per sq.mm for cpu and 40 for GPus and this is 10nm, not 7nm. Intel 14nm is only 37 M transistors per sq.mm. so what.

 

[Fluffmeister]

Another market they could do with a third player, hey ho.

 

[Patriot]

The point is, they have real work to do
Would you dare someone without a PhD in CS to pick between an i5-1035G7 and an i7-10510U?

I would definitely dare someone WITH a PhD in CS as pure software engineers know fuck all about hardware AND intel's line is an utter clusterfuck for hardware engineers.

 

[notb]

The point is, they have real work to do

Why exactly is updating products not real work?

Would you dare someone without a PhD in CS to pick between an i5-1035G7 and an i7-10510U?

Honestly, I don't know what you meant here.
As @Patriot already said: Computer Science isn't concerned with chip design, since computers are not devices made of semiconductor chips. Computer is an abstract machine that can perform a sequence of instructions (called "a program").