Discussion Intel current and future Lakes & Rapids thread

Page 783 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Hitman928

Diamond Member
Apr 15, 2012
6,695
12,370
136
But overall area also decreased, along with the number of transistors. The growth of the gate pitch might have been minor, or could have been major, but overall if the % increase in area decreasing outpaced the %# transistor, density would grow.

You are arguing that you can increase the distance between transistors (decreasing density) but putting less transistors means transistor density increases. Please show me how the math works out on that.
 
  • Haha
Reactions: gavitk

Joe NYC

Diamond Member
Jun 26, 2021
3,655
5,199
136
I agree it makes for an awkward lineup and branding, but Intel's only alternative would be to launch 14th gen and 15th gen desktop simultaneously, which would be confusing in its own right. My bet is we'll see 14th gen launch as a combo of MTL mobile parts and RPL refreshes for filler, and then 15th gen as LNL (premium mobile), ARL (mobile and higher end desktop), and MTL (low end desktop).

Intel has nothing to worry about from AMD re: awkward branding, after seeing AMD notebook model numbers.
 
Last edited:

Geddagod

Golden Member
Dec 28, 2021
1,531
1,625
106
You are arguing that you can increase the distance between transistors (decreasing density) but putting less transistors means transistor density increases. Please show me how the math works out on that.
1679257273175.png
This slide from Intel 10SF literarily says improved transistor density.
 
  • Like
Reactions: Henry swagger

Hitman928

Diamond Member
Apr 15, 2012
6,695
12,370
136
View attachment 78405
For example, this slide literally states improved transistor density from COAG.

Yes, COAG allows you to decrease the cell height and increase density if the design allows for it. However, this was introduced with the original 10 nm in Cannon Lake and was part of their density increase claim over their 14 nm process.

This slide from Intel 10SF literarily says improved transistor density.

This was not new for 10SF. It was in Cannon Lake.
 

Geddagod

Golden Member
Dec 28, 2021
1,531
1,625
106
Yes, COAG allows you to decrease the cell height and increase density if the design allows for it. However, this was introduced with the original 10 nm in Cannon Lake and was part of their density increase claim over their 14 nm process.
I think it's improvements to COAG. Afterall, 10SF didn't just deal with gate pitch.
Gate pitch isn't everything...
 

Geddagod

Golden Member
Dec 28, 2021
1,531
1,625
106
Quoting Intel;
"In an email to IT World Canada, Intel explained that while the gate pitch impacts the size of the standard cell, it’s only a small part of the various factors that affect density. Design choices, metal routing layers, and many other factors all determine transistor density. Additionally, Intel said that “On the whole, our 10nm SuperFin technology is equivalent in density to foundry 7nm processes.”
 
  • Like
Reactions: Henry swagger

Hitman928

Diamond Member
Apr 15, 2012
6,695
12,370
136
I think it's improvements to COAG. Afterall, 10SF didn't just deal with gate pitch.
Gate pitch isn't everything...

What improvements could you add to COAG to allow for higher density that would allow you to increase density?

Gate pitch isn't everything. For example, there's required dummy gates (which Intel decreased but again, was introduced with the original 10 nm process). However, there is no indication that Intel did anything except decrease density.
 

Geddagod

Golden Member
Dec 28, 2021
1,531
1,625
106
What improvements could you add to COAG to allow for higher density that would allow you to increase density?

Gate pitch isn't everything. For example, there's required dummy gates (which Intel decreased but again, was introduced with the original 10 nm process). However, there is no indication that Intel did anything except decrease density.
There is no indication Intel decreased density.
They increased gate pitch, but also worked on other parts of the transistor.
I admitted I was wrong that Intel increased density with SF, I obviously misread that quote
But saying Intel decreased density is wrong as well.
Intel claimed with OG 10nm that they were equivalent to external 7nm
And they continued to claim they were equivalent to external 7nm after SF as well (in density).
 

Hitman928

Diamond Member
Apr 15, 2012
6,695
12,370
136
Quoting Intel;
"In an email to IT World Canada, Intel explained that while the gate pitch impacts the size of the standard cell, it’s only a small part of the various factors that affect density. Design choices, metal routing layers, and many other factors all determine transistor density. Additionally, Intel said that “On the whole, our 10nm SuperFin technology is equivalent in density to foundry 7nm processes.”

I don't think this quote is supporting your stance like you think it is. Remember, Intel originally claimed higher density for 10 nm compared to foundry 7 nm processes. Now they are saying it is, "equivalent".
 
Last edited:

Hitman928

Diamond Member
Apr 15, 2012
6,695
12,370
136
There is no indication Intel decreased density.
They increased gate pitch, but also worked on other parts of the transistor.
I admitted I was wrong that Intel increased density with SF, I obviously misread that quote
But saying Intel decreased density is wrong as well.
Intel claimed with OG 10nm that they were equivalent to external 7nm
And they continued to claim they were equivalent to external 7nm after SF as well (in density).

Originally Intel claimed over 100 MTr/mm2, which was the highest in the world at the time.

Here is an article analyzing the various nodes from different foundries. They give this one line comparison between 10nm and 10nm+

Ice Lake is based on Intel’s 10nm+ Node and its density is significantly lower than that of 10nm used for Cannonlake.


Also, just to address the following since you never actually went back and tried a realistic calculation:

A scenario to show case this:
100 transistors in 10 mm^2, 50 of them are buffers , 10 transistors per mm^2
80 transistors in 8 mm^2, 30 of them are buffers, 10 transistors per mm^2
Density is number of transistors and area.

In your example, if you increase gate pitch, thus decreasing density say 10%, you would get 80 transistors in 8.8 mm2. Your transistor density would drop to 9 transistors per mm2 but your overall area would still go down 12%. Thus, decreasing density leading to using less buffers actually allows you to use less area overall which is what the Anandtech article was saying.

I don't think you will find any real analysis that comes to the opposite conclusion. With that, I'm really out for this topic. Like I said, believe what you like, I don't care. We'll see how/when Intel 4 and 20a land.
 

Geddagod

Golden Member
Dec 28, 2021
1,531
1,625
106
I don't think this quote is supporting your stance like you think it is. Remember, Intel originally claimed shigher density for 10 nm compared to foundry 7 nm processes. Now they are saying it is, "equivalent".
Apparently Intel was claiming their 10nm was better than TSMC's 7nm as late as 2021, and they would have been on SF at that point (not yet ESF either)
And I don't think Intel ever said their 10nm was "better" than TSMC 7nm, despite their original density numbers being reported to be <10% higher
But by the time that quote rolled around in 202, TSMC 7nm with EUV surpassed Intel 10nm density by <10% as well for HD cells
 

Geddagod

Golden Member
Dec 28, 2021
1,531
1,625
106
I don't think you will find any real analysis that comes to the opposite conclusion. With that, I'm really out for this topic. Like I said, believe what you like, I don't care. We'll see how/when Intel 4 and 20a land.
The problem is there literarily is no real analysis that indicates that Intel 10SF decreased density, at the very least by no significant margin,
since Intel has consistently claimed their 10nm was as good as external 7nm, before or after SF, for density.
Gate pitch might have increased sure, but Intel also claimed they worked on numerous other parts of the transistor as well.
But ye, we will see where Intel 4 and Intel 20A land...;)
 

Hulk

Diamond Member
Oct 9, 1999
5,140
3,728
136
If Intel plans on using all of the even numbers below 20 they'll have a whole lot of "new" nodes that used to be "+" nodes. Good for them to bend reality as they see fit.
We're all living in the Matrix anyway.
 
  • Haha
Reactions: Tlh97 and Geddagod

Exist50

Platinum Member
Aug 18, 2016
2,452
3,106
136
But then I predict that Intel (as usual) will fall in love with the yields that are finally occurring with Intel 4 and 15th generation in 2025 will be a Meteor Lake Refresh that include better, lower voltage mobile parts as well as high clock speed desktop parts.
By 2025, MTL should only make sense for very cost sensitive markets. They should have ARL on 20A to fill in for mobile, and maybe we'll even see it in desktop eventually? And if nothing else, they will want to push 20A ARL out in as high a volume as possible as a proof point for their fabs. And hopefully by the end of 2025, we see something (Cougar Cove?) on 18A across the board.
You would have to wonder what the difference between the I9 and I5 skus are going to be then, if ARL is i7 and i9, and MTL is i5 and below.
ARL also has a secret that will make the SKU choices even more interesting. But regardless, I think they'll just have to live with the generational gap in performance between the i5 and i7. It'll be a mess, but unless they introduce a downmarket ARL die, they don't have much of a choice. Hopefully that's temporary, however, because even just from a DIY perspective, there won't be a value champ like the 12400 from that lineup.
 

Geddagod

Golden Member
Dec 28, 2021
1,531
1,625
106
An Amateur's Project:
A fun mockup of the EMR core (GLC++? haha) based on 5MB L3 EMR rumors:
1679267125480.png
On the left we have the mockup of a 5MB L3 EMR core, on the right we have the current SPR core.
Method:
So the current SPR has 1.875MB of L3 per core. This is divided into 2 L3 cache arrays, which are slightly differently sized. By adding the two areas together, and then dividing 1.875/total area, we can find MB/area.
Multiply this by the area of each of the data arrays to find how much each array holds, and we get how much cache each data array holds.
I then replicated the blocks in a method that seemed area efficient and continued to hold a 'square' or 'rectangle' shape for the core, as well as leaving space for extra control for the cache.
1679267934776.png
Conclusion for core:
It looks like the core size grew by ~20%. While there is some extra space in between some of the blocks, I probably underestimated the amount of space needed for L3 control as well as Mesh Agent, as I suspect that would probably grow a bit in EMR from the extra L3 cache and in order for that to better communicate between cores.
What about the CPU?
The "core" part of one of the SPR CPU tiles (including the memory control tile) is only slightly larger than 50% of the tile.
1679268890550.png
Increasing that by 20% means that the total tile is only ~10% larger than the SPR tile. For more than 2.5X the L3 cache per core, an only ~10% additional die size does not sound that bad. Especially since, from chips and cheese testing, SPR's major bottleneck seems to be the L3 cache.
64 cores?
If we look at the core configurations for SPR, we see a 4 x 4 array of cores, with a memory control tile in the mix. For SPR, the max is 60 cores, so for EMR reaching 64 cores, the configuration of the cores have to have been changed.
I see two options here: a 4 x 4 array of cores, with the memory control tile moved off the array closer to where the DDR5 PHY or Accerelators and PCIE or CXL blocks are or....
a new configuration. 5 x 4? 6 x 3?
Here's a mockup of a potential 5 x 4 configuration:
1679270296678.png
The problem with Intel implementing a potential new configuration of cores is that it would most likely take longer to design, and have to disable 2 or 3 cores (if it stays on one memory controller per tile) for the 64 core max rumors to be true. The Tile also grows by closer to 15-20%.

So ye, that's pretty much it :)
 

Exist50

Platinum Member
Aug 18, 2016
2,452
3,106
136
64 cores?
If we look at the core configurations for SPR, we see a 4 x 4 array of cores, with a memory control tile in the mix. For SPR, the max is 60 cores, so for EMR reaching 64 cores, the configuration of the cores have to have been changed.
I see two options here: a 4 x 4 array of cores, with the memory control tile moved off the array closer to where the DDR5 PHY or Accerelators and PCIE or CXL blocks are or....
a new configuration. 5 x 4? 6 x 3?
Don't be afraid to think outside the box a little ;). Literally, in this case.
 

Hulk

Diamond Member
Oct 9, 1999
5,140
3,728
136
By 2025, MTL should only make sense for very cost sensitive markets. They should have ARL on 20A to fill in for mobile, and maybe we'll even see it in desktop eventually? And if nothing else, they will want to push 20A ARL out in as high a volume as possible as a proof point for their fabs. And hopefully by the end of 2025, we see something (Cougar Cove?) on 18A across the board.

Based on Intel's recent history with 14nm and 10nm I don't think they do just 1 generation on Intel 4 for a couple reasons.

First, I would be astounded if ARL was ready to go on the desktop in 2025 on 20A. 20A will hold them up.

Second, in order to avoid a Rocket Lake fiasco (backport to Intel 4 for ARL) Intel will keep ARL on 20A, this means when 20A inevitably falls behind schedule that will leave them with MTL refresh on a tweaked Intel 4, maybe they'll rename it "Intel 3. Maybe that's whey they saved numbers between Intel 4 and 20A.

So what I'm predicting is Raptor Refresh and maybe MTL mobile 2023. MTL desktop and mobile (again) on the refined Intel 4 desktop node in 2024.

ARW 20A mobile only 2025.
 

Exist50

Platinum Member
Aug 18, 2016
2,452
3,106
136
that will leave them with MTL refresh on a tweaked Intel 4, maybe they'll rename it "Intel 3. Maybe that's whey they saved numbers between Intel 4 and 20A
Uh, Intel 3 is officially a thing...
ARW 20A mobile only 2025.
I think it's reasonably likely that we only see ARL-P 20A devices on shelves in 2025, but they should hopefully be able to move quickly and fill out the lineup. Maybe have some 18A chips in time for the holidays.
 
  • Like
Reactions: IntelUser2000

DrMrLordX

Lifer
Apr 27, 2000
22,908
12,979
136
assuming Sierra Forrest is cancelled.

I don't think even MLID went so far as to claim that it was cancelled entirely. His claim (which can not be verified, and may never be verified due to the nature of the product) was that the high core-count variants of Sierra Forest (AP) were gone, while the lower-core count versions were being forced out the door ASAP to fill the orders of a select customer "on time" (by Q2 2023, if I recall the leak correctly. Honestly I don't want to go back down his rabbit hole to look).
 
  • Like
Reactions: HurleyBird

Exist50

Platinum Member
Aug 18, 2016
2,452
3,106
136
It's old. SRF-AP was canned ages ago.
I wouldn't be quite so sure about that. There's the rumor it was revived as a sort of dual-SP die setup. And looking at that image, the copyright is from 2023. Can't be that old.
 

Exist50

Platinum Member
Aug 18, 2016
2,452
3,106
136
View attachment 78398
Anyone know why Harukaze on twitter deleted this? It certainly looks legit.
Also 500 watts on GNR vs 600 watts on Turin?
What's the difference between Intel and AMD TDP again?
I certainly think this is legit. Lots of interesting details to dissect.
  • Up to DDR5-6400 for 1 DIMM per channel.
  • MCR support confirmed, up to an effective 8000MT/s (hope this eventually trickles down to HEDT...)
  • Self-Boot == no chipset required?
  • Goes up to at least 500W.
Also, wow, 20 layer PCB. That must cost a pretty penny.
 
  • Wow
Reactions: wilds

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,787
136
This new naming structure allows them to better align with TMSC node "sizes" and allows them to get away from the dreaded "+" syndrome. ie Intel 20A+ will not exist and actually be Intel 18A.

I don't fully agree with this either. Look at this slide.

The plusses are just that, nice improvements but small. 5.5%, 3.8%, 5.8%, 5.9%. And it's not multiplicative either, it's additive. 10nm SF wasn't called a + because it single handedly brought improvements equal to four plusses.

Let's look at Intel 3 and 18A shall we?

Intel 4 - 20%
Intel 3 - 18%
Intel 20A - 15%
Intel 18A - 10%

The performance gains are big. Interestingly in this aspect 20A is LESS than the Intel 3 process. More work for less gains, this is the slow death of Moore's Law.

While we heavily focus on density, performance is very important. TSMC's 20nm brought great density gains, but almost no one used it, and they waited for 16nm with FinFET. Likewise TSMC's N2 brings barely any density gain, but performance seems good. Probably is going to be similar with Intel's 20A. It's about performance not density. We used to get both readily but that ship is slowly sailing away.

And you are also forgetting that Intel 3 is being skipped on client, but Intel 4 is being skipped on server. That's why there's a gap. They aren't changing names, they are skipping processes.

To start of with, TSMC doesn't disclose HCC SRAM density
Intel 7/Intel 10nm is ~7nm.
Intel 10nm HD is ~ 7nm
Intel 7 HP ~ TSMC 7nm
Intel 7 is roughly 15% larger than TSMC 7nm HDC cells

You got a lot to learn buddy. Thinking buffers aren't made of transistors haha. Buffers are basic building blocks. Not logic, but important.

If anything, the SF and ESF(called Intel 7 now) are less dense than 10nm in Icelake. The clock speed increase didn't come for free. Yes, pitches did increase. And pitches are one thing that directly affects density, unlike things like COAG which is a maybe.

We know the whole 14nm/10nm density claim was quite misleading since while it could be true, it did not apply to their Core lines. Atom gained awesomely and so did their GPUs. But that's not bringing them much money do they? They tried to do a mobile shift in 14nm, but they were in essence killing the hen that lays golden eggs(just like the story to get more of them inside her).

ARL also has a secret that will make the SKU choices even more interesting.

Very interesting. About ARL+MTL, little bit of price adjustments and plus or minus 100MHz adjustments and P+E core ratios can easily make it work.

@Hulk If they do as you planned, then they'll lose on desktop again. I personally am not that pessimistic. They know that Arrowlake, even in small volumes at the high end is necessary to charge extra at the high end, because that's where the margins and profits are. If Meteorlake is at top, then you end up at x600K, and maybe x700K at max. That's a loss.
 
Last edited: