• Guest, The rules for the P & N subforum have been updated to prohibit "ad hominem" or personal attacks against other posters. See the full details in the post "Politics and News Rules & Guidelines."

Discussion Intel current and future Lakes & Rapids thread

Page 439 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Hougy

Member
Jan 13, 2021
32
39
51

Ajay

Diamond Member
Jan 8, 2001
9,482
3,950
136
Looks like full 2D mesh with router at each node? Just like SKL-SP imo. L3 cache size has grown and hopefully mesh interconnect is wider too. I think it is 99% safe to assume it is same architecture but iterated on.
Skylake mesh was shown as "rings" as well though:



The difference is that in the tweeted image 1) the interconnects extend beyond the node to the border of the die and 2) there are additional interconnect "rings" at the top and right edge of it.

The big question is are those actual topology design changes worth mentioning beyond PR, or are these just changes to the visual presentation?
Hmm, just my mind playing tricks on me I guess. Intel just drew the block diagram a bit differently <shrug>.
 

Accord99

Platinum Member
Jul 2, 2001
2,214
101
106
If I understand the current situation, Tiger Lake-U cannot be undervolted in software, but Tiger Lake-H can.
A redditor received their 11800H based laptop and was able to undervolt and overclock, -70mV with 4.4 GHz all-core with Cinebench R23.

This run was done with max fans. In high performance and overclocked on the CPU by 200Mhz. So it ran at 4.4ghz. The CPU can be Overclocked +4 bins. So max single core would be 5ghz and max all core is 4.6ghz. In order to achieve that I need to increase the voltage rather than undervolt and I would need to use liquid metal to cool it. For now I’m happy at 4.4ghz all cores with an undervolt of -70mV.

At stock with voltage tuned -75mV it was pulling aroubd 70w package sustained and sustained 4.2ghz on all cores and scores 13.8K multi in R23.


https://www.reddit.com/r/intel/comments/nfmytu
 

IntelUser2000

Elite Member
Oct 14, 2003
7,608
2,469
136
Except the 10++nm process listed there is 10SF not ESF (they shifted the names soon after Cannon Lake)
If Alderlake ends up having lower max clock, that's because Skylake has been known by Intel engineers for 6 years while Golden Cove is two generations newer than that.

Clock-wise the lower power consumption of 10nm will also contribute to a potentially higher clock if thermals are the limit. I doubt it but it'll be things other than process that'll determine clocks for Alderlake.

Hardware Unboxed tested TGL H45 and looked at performance with different TDP settings and boost disabled. This is a missing piece of the puzzle which shows that TGL H really likes more power, as increasing TDP towards desktop levels helps TGL catch up in terms of efficiency.

It's almost as if TGL H45 was meant to be TGL-S :p

View attachment 44675
The graphs are very interesting. Typically processes aim for a steeper graph on the low end that tails off on higher power, which disproportionately benefits low power CPUs. The steeper curve was also pronounced especially with transition to FinFET processes.

I think maybe it's showing that Intel has rejigged their process in a way that's different from their 22nm and 14nm processes.* Since the SF variants are targetting CPUs with high clocks, it makes sense.

At the lower power Tigerlake chips retain the advantages over their 14nm(it should as smaller processes save power due to low capacitance) so they've improved clock/power efficiency at all points.

*One might remember the disappointing clocks of their 22nm Ivy Bridge desktop chip. I've called them out saying they sacrificed their bread and butter PC CPUs to benefit nascent market for them - mobile Atoms. The claims of 37% gain using 22nm happened to apply exactly to the Silvermont Atom chips.

Previously Intel would have only adjusted core and design philosphies. 22nm marked the time when Intel changed their core strength(process) to hope to compete in mobile. It was of course too much for them at that time.

I also believe it was the over-focus on the mobile market that caused the delay to their 14nm and 10nm process. Had they stayed with PC as a focus on 22nm, they might have been in a much better position to keep the lead.

The lesson here is:
You do not bet the farm on something new and of questionable value to you, no matter how much others scream that it's of dire importance. You have to treat it as an extension.
 
Last edited:

JoeRambo

Golden Member
Jun 13, 2013
1,246
1,183
136
They made some very unfortunate decisions to cut R&D in ~2012-2015 timeframe, and it made completely no sense for them at the time to focus on core business of Server+Desktop+Laptop cpus, as they were completely dominating. So decisions were made to go after mobile and foundry, and that meant different tuning for processes and diabolical hunt on density to beat other foundries.
With hindsight i think all that quad patterning madness was of long term play of becoming a foundry for Apple, first modems and then SoCs.

Results of decisions during that time frame has put the very survival of Intel in question. If they had more R&D money, things might have played out differently, maybe two competing processes (one tuned for core business), maybe "backup" process with heavier use of EUV.

Alder Lake and Saphire Rapids are gonna be most important products for Intel ever (since C2D). if Golden Cove is underwhelming, or Saphire Rapids servlet level architecture flops, it is pretty much game over for them. Only political tides can save them if they flop.
 

eek2121

Golden Member
Aug 2, 2005
1,290
1,370
136
People are predicting lower clocks for Alder Lake S than Rocket Lake, and they might be right, but I just remembered that there were old Intel slides predicting 10nm+++ to have higher performance than 14nm+++. I think they were from when they showed a roadmap with a new node every two years for this decade
Current known ES chips are up to 4.7 Ghz. I suspect we will see little, if any clock regression in the final product.

A redditor received their 11800H based laptop and was able to undervolt and overclock, -70mV with 4.4 GHz all-core with Cinebench R23.




https://www.reddit.com/r/intel/comments/nfmytu
I can’t wait to get mine. Not shipping until next month.
 

ondma

Platinum Member
Mar 18, 2018
2,317
927
106
They made some very unfortunate decisions to cut R&D in ~2012-2015 timeframe, and it made completely no sense for them at the time to focus on core business of Server+Desktop+Laptop cpus, as they were completely dominating. So decisions were made to go after mobile and foundry, and that meant different tuning for processes and diabolical hunt on density to beat other foundries.
With hindsight i think all that quad patterning madness was of long term play of becoming a foundry for Apple, first modems and then SoCs.

Results of decisions during that time frame has put the very survival of Intel in question. If they had more R&D money, things might have played out differently, maybe two competing processes (one tuned for core business), maybe "backup" process with heavier use of EUV.

Alder Lake and Saphire Rapids are gonna be most important products for Intel ever (since C2D). if Golden Cove is underwhelming, or Saphire Rapids servlet level architecture flops, it is pretty much game over for them. Only political tides can save them if they flop.
Not sure what you mean by "game over". Despite their problems, Intel still has the majority of the market, and AMD simply does not have the capacity to supply the market alone. So I dont think Intel is in any way close to going out of business. OTOH, they have lost the performance crown, and I dont see them regaining it in desktop or server. If Alder Lake concept pans out, I could see them being very competitive if not leading edge in mobile. They should take the lead in IPC with the big cores, and could tune them to be more efficient, letting the small cores add multi threaded performance. They really need a ground up big core redesign to make it smaller and more efficient though.
 
  • Like
Reactions: Tlh97 and ryan20fun

dmens

Platinum Member
Mar 18, 2005
2,237
819
136
If Alder Lake concept pans out, I could see them being very competitive if not leading edge in mobile. They should take the lead in IPC with the big cores, and could tune them to be more efficient, letting the small cores add multi threaded performance. They really need a ground up big core redesign to make it smaller and more efficient though.
Hah. That boat sailed permanently with Apple M1. It is already over.
 

dmens

Platinum Member
Mar 18, 2005
2,237
819
136
There's only so much of the market Apple can get. After all there's a ton of people who are hopelessly tied to Windows.
That could well change when more people realize the massive performance difference between apple silicon and intel space heaters.
 

uzzi38

Golden Member
Oct 16, 2019
1,817
3,654
116

dmens

Platinum Member
Mar 18, 2005
2,237
819
136
It sailed, yes, for people who are comfortable with using Apple (-approved) software every single day. In the average PC and Notebook space nothing is over, regardless of how much people hype M1.
That's fine. The boat that sailed is Intel being the leading edge in mobile (or anything else for that matter).
 

jpiniero

Lifer
Oct 1, 2010
10,105
2,364
136

DrMrLordX

Lifer
Apr 27, 2000
17,840
6,816
136
Not sure what you mean by "game over". Despite their problems, Intel still has the majority of the market, and AMD simply does not have the capacity to supply the market alone.
You did see Mizuho's estimates for Intel's 7nm capacity, no?
 

Asterox

Senior member
May 15, 2012
632
985
136
People are predicting lower clocks for Alder Lake S than Rocket Lake, and they might be right, but I just remembered that there were old Intel slides predicting 10nm+++ to have higher performance than 14nm+++. I think they were from when they showed a roadmap with a new node every two years for this decade
Slides, huh they can hug absolutely anything.For example, this is old Intels jewel of future progress.

OMq3C3I (1).png
 
  • Like
Reactions: rUmX

eek2121

Golden Member
Aug 2, 2005
1,290
1,370
136
Hah. That boat sailed permanently with Apple M1. It is already over.
That could well change when more people realize the massive performance difference between apple silicon and intel space heaters.
Apple will always be a minority player. They barely compete in the desktop/laptop space (10% marketshare on a good day) and don’t really compete at all in the server space (well under 0.5% marketshare)

Having a slightly faster CPU won’t fix that.

Cezannne and Tiger Lake certainly are not behind significantly in any metric, despite both chips being a node behind.

I love the inclusion of an AVX-512 benchmark to just casually inflate the overall IPC uplift with a +78% result (removing it shows a 16.71% IPC uplift on average) and the inclusion of not one, not two, but 3 different tile-based renderers (Cinebench R15, R20 and R23).
AVX-512 is a valid instruction set in cases where it is used. I am not claiming that is the case here, but if I had a software package I used regularly that was 78% (or even 40%) faster on an Intel vs. AMD system, I would have to go with the Intel system.

Luckily, I do not.
 

dmens

Platinum Member
Mar 18, 2005
2,237
819
136
Apple will always be a minority player. They barely compete in the desktop/laptop space (10% marketshare on a good day) and don’t really compete at all in the server space (well under 0.5% marketshare)

Having a slightly faster CPU won’t fix that.
LOL. It is energy efficiency that matters, especially in mobile. In that metric, apple silicon is somewhat more than "slightly faster". Also, nice attempt at moving the goalposts by citing market share as a replacement for technological leading edge.
 

Ajay

Diamond Member
Jan 8, 2001
9,482
3,950
136
From what I gather he meant that in a strictly technological sense.
He pointed out that Intel doesn’t have many EUV lithography machine, especially relative to the number of wafers they manufacture every month. It true, it’s another sign of gross mismanagement. The only other scenario that makes sense to me is that 7nm is going to be a short node and Intel will be relying on 10/7nm 3D hybrid packaging. Intel would have to go all in on their 5nm. I don’t see away for Intel to get back on a leading edge process in the next 10 years, if ever. That’s probably why the US semiconductor funding initiative is focused mainly on R&D.
 

ASK THE COMMUNITY