Discussion Intel current and future Lakes & Rapids thread

Page 231 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

coercitiv

Diamond Member
Jan 24, 2014
6,151
11,686
136
Frequency hasn't really been the problem, it's always been about yield. Of course if you are actually intending to sell the product, you need to make sure you get enough chips that can pass the validation.
Are you saying their parametric yields were fine but their functional yields were in the gutter?
 

mikk

Diamond Member
May 15, 2012
4,112
2,108
136
The problem was clearly both clock and yield issue. I mean Intel expected themselves from the beginning that the first 10(+) version has a much worse performance than 14++ and only the second 10+(+) is on par with 14++, as a result ICL-U singlethread wasn't better than CML-U. Possibly there is still a small performance gap but It looks quite good for TGL-H when TGL-U can boost up to 4.8 Ghz even if it's a TVB boost, we should see a 5.0 Ghz boost next year.
 
  • Like
Reactions: french toast

dullard

Elite Member
May 21, 2001
24,998
3,327
126
The problem was clearly both clock and yield issue. I mean Intel expected themselves from the beginning that the first 10(+) version has a much worse performance than 14++ and only the second 10+(+) is on par with 14++, as a result ICL-U singlethread wasn't better than CML-U. Possibly there is still a small performance gap but It looks quite good for TGL-H when TGL-U can boost up to 4.8 Ghz even if it's a TVB boost, we should see a 5.0 Ghz boost next year.
This article, over 3 years old, is still quite relevant: https://www.pcmag.com/news/intels-10nm-process-its-more-than-just-chip-scaling:
(I did find it interesting—and a little worrisome—that, although these charts show the 10nm nodes clearly requiring less power than the 14nm nodes, they suggest that the first 10nm nodes will not offer as much performance as the latest 14nm ones.)

However, isn't the problem also possibly a reliability issue? A theoretical processor can have all the speed and yield you could want, but if it isn't reliable then it is useless. Cobalt layers are new and unproven. Do the cobalt layers bring their own electromigration issues, their own thermal expansion issues, their own RC issues (with 4x the resistivity of copper), etc.?
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
Interesting article.
I doubt the 15% better performance of 10nm++ vs 14nm++ will ever happen.

Depends on the clock speed the product is targetting.

Zero chance - Peak frequency for laptops and desktops. Peak frequency is limited by thermals not transistor performance.

Quite good - GPUs/CPUs running at <3GHz
 

french toast

Senior member
Feb 22, 2017
988
825
136
Well 10nm is now bearing fruit for Intel, Tiger lake is is shaping out to be a solid product.

If they managed to get 16 x willow Cove cores on desktop using EMIB (2x8) on this process they would be extremely competitive with any Zen 3 implementation AMD could muster.

If Tigerlake H can really boost to 5ghz in mobile form factor early next year I doubt even cezzanne would beat it in St perf.
Still AMD is already on 8 cores, Vega 8 is very Conservative on tech and performance and I wouldn't bet against it matching Tigerlake in gaming.

Anyway great competition.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
@french toast The problem is leaks are showing Alderlake being an 8+8 product.

If they chose to, they could make a version of Alderlake with 2x 8+8 cores. But 8 Golden Cove cores are already very large on 10nm, and realistically I think they need 7nm to do so. 2x 180mm2 dies are really pushing it, and it'll need an GMIO(Graphics, Memory, IO) die on top of that. Besides, they'll lose a lot of memory latency advantages by moving to a chiplet unlike AMD.

If Xe iGPU is fast as they claim then it'll beat Renoir significantly in graphics(30-50%).

But I'm not confident of their GPU team. See I don't think its just drivers. We initially blamed graphics for X3000, but it just had crappy geometry hardware and occlusion capabilities. Those are merely high-level features they lacked. Now imagine what low level details are behind/missing on Gen 11, or possibly even Gen 12?

Whether its lack of proper leadership, company culture or experience, they are 3rd tier. It's not like in CPUs where they faltered once but came back. They NEVER had leadership in graphics.
 
Last edited:

LightningZ71

Golden Member
Mar 10, 2017
1,627
1,898
136
The problem for Intel with Xe, even if they can execute according to their plan, is that they won't be competing against Vega8 as it exists on Renoir. It'll be up against whatever configuration AMD uses on Cezanne. While we don't expect Cezanne to have dramatically different iGPU characteristics, it will still have at least some improvements, even if those are more process related. I suspect that AMD will take advantage of whatever benifits that the more mature N7(+) process has by then to not only implement the better Zen 3 CPU cores, but to also likely be able to fit in another 2-4 CUs in the iGPU section. Assuming that they can achieve at least another 10% bump in iGPU clock speed, and at least include 25-50% more CUs, coupled with a likely modest step in memory speeds (DDR4-3600/LPDDR4X-4400+), they'll easily be able to offer another 20% in measurable gaming iGPU performance without making any dramatic changes. And, within 6 months of Intel's product release, there'll be a successor on N5P, with RDNA CUs, with DDR5, that should have rather dramatically better performance than anything AMD has out today in an iGPU. So, at best, I see Intel managing parity with AMD in the APU space for a while at the top end.
 
  • Like
Reactions: spursindonesia

Cardyak

Member
Sep 12, 2018
72
159
106
Tom at MLID has receive some leaks about Intel's future CPU road map. He's not dug through it yet from what I can make out and as always you need to take these videos with massive amounts of salt. But in this podcast he did let slip that Intel seems to have a future CPU Design dubbed "Lunar"


So if this is to be believed, is the assumed road map now: Ice Lake -> Tiger Lake -> Alder Lake -> Meteor Lake -> Lunar Lake?
 

french toast

Senior member
Feb 22, 2017
988
825
136
@french toast The problem is leaks are showing Alderlake being an 8+8 product.

If they chose to, they could make a version of Alderlake with 2x 8+8 cores. But 8 Golden Cove cores are already very large on 10nm, and realistically I think they need 7nm to do so. 2x 180mm2 dies are really pushing it, and it'll need an GMIO(Graphics, Memory, IO) die on top of that. Besides, they'll lose a lot of memory latency advantages by moving to a chiplet unlike AMD.

If Xe iGPU is fast as they claim then it'll beat Renoir significantly in graphics(30-50%).

But I'm not confident of their GPU team. See I don't think its just drivers. We initially blamed graphics for X3000, but it just had crappy geometry hardware and occlusion capabilities. Those are merely high-level features they lacked. Now imagine what low level details are behind/missing on Gen 11, or possibly even Gen 12?

Whether its lack of proper leadership, company culture or experience, they are 3rd tier. It's not like in CPUs where they faltered once but came back. They NEVER had leadership in graphics.
Yea I don't trust their drivers or lower level optimisation, I think it will get better with gen 12..but I still think it will under punch for its weight compared to Vega 8...in real games at least.

Eventually they will catch up enough, they keep throwing money at the issue and grabbing engineers, over time the gap will be small.

I honestly think Intel's 10nm +(+) process is at the very LEAST equal to TSMC N7Pin perf/watt- if not better, a giant core like willow Cove with heaps of cache and AVX 512 Floating point units pushing to 4.8ghz in a 15w 'tdp'?.. that's very impressive indeed (if true.)

Now yields and density are firmly in TSMC's corner with N7P, else we wouldn't be stuck on 4 cores and no desktop 10nm.
 

lobz

Platinum Member
Feb 10, 2017
2,057
2,856
136
Yea I don't trust their drivers or lower level optimisation, I think it will get better with gen 12..but I still think it will under punch for its weight compared to Vega 8...in real games at least.

Eventually they will catch up enough, they keep throwing money at the issue and grabbing engineers, over time the gap will be small.

I honestly think Intel's 10nm +(+) process is at the very LEAST equal to TSMC N7Pin perf/watt- if not better, a giant core like willow Cove with heaps of cache and AVX 512 Floating point units pushing to 4.8ghz in a 15w 'tdp'?.. that's very impressive indeed (if true.)

Now yields and density are firmly in TSMC's corner with N7P, else we wouldn't be stuck on 4 cores and no desktop 10nm.
That clockspeed may only be attainable (for longer than 3 milliseconds) in the 28W TDP configuration.
 
  • Like
Reactions: spursindonesia

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
Eventually they will catch up enough, they keep throwing money at the issue and grabbing engineers, over time the gap will be small.

Intel has lot of money, but their execution sucks, and is not consistent. Also they have a lot of drama.

Some articles were saying that the fault is due to their culture, and its a result of the influence of its most significant CEO, Andy Grove. It points out while such culture worked under him, it doesn't with others. When a CEO is such a superstar I think the rest of the company starts droning on, and when he leaves it becomes directionless.

I honestly think Intel's 10nm +(+) process is at the very LEAST equal to TSMC N7Pin perf/watt- if not better, a giant core like willow Cove with heaps of cache and AVX 512 Floating point units pushing to 4.8ghz in a 15w 'tdp'?.. that's very impressive indeed (if true.)

For a single core its not much of a perf/watt issue. U chips have top frequencies not different from 95W desktop chips.
 

french toast

Senior member
Feb 22, 2017
988
825
136
Intel has lot of money, but their execution sucks, and is not consistent. Also they have a lot of drama.

Some articles were saying that the fault is due to their culture, and its a result of the influence of its most significant CEO, Andy Grove. It points out while such culture worked under him, it doesn't with others. When a CEO is such a superstar I think the rest of the company starts droning on, and when he leaves it becomes directionless.



For a single core its not much of a perf/watt issue. U chips have top frequencies not different from 95W desktop chips.
Intel reminds me of old Nokia in some respects, although Intel has a ha bit of making silly money even when they are not on their game.
Yea defo had a culture problem, whether they can fix that long term is anyone's guess.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
Intel reminds me of old Nokia in some respects, although Intel has a ha bit of making silly money even when they are not on their game.

Intel unlike Nokia is in a much more profitable business, and is little more diversified.

I think they should open up a bit more and not make everything so much for themselves. That'll prevent the whole thing from going under when one section falters badly.

I know their WiFi division is very successful. It's because they have a discrete version and works in AMD devices. Practically all modern desktop/laptop WiFI is Intel.

But that's very limited in terms of revenue.

They need to open up Optane DIMMs to work on all platforms. The false idea that monopoly companies have is that they think having a revolutionary tech closed to themselves will help them. In reality and over long term it backfires. Nascent technologies need every help it can get.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
It's a great improvement compared to not sucking, I completely agree.

Icelake's parametric yield sucking seems to be reflected in battery life tests. It's super good in idle, far better than Whiskeylake. Yet the battery life is equal at most and generally worse.

So the chip needs to use more current and voltage to be at the same performance, nevermind frequency as the 14nm predecessor.

GPU seems to be ok, so its at higher, CPU-level clocks where it all runs out of control.
 

Ajay

Lifer
Jan 8, 2001
15,332
7,792
136

RetroZombie

Senior member
Nov 5, 2019
464
386
96
It looks like Jim Keller has resigned from Intel:
In he's last interview it seams he was giving some 'internal messages' like that one of the how to manage people, and the one of that an design only should lasts for 5 years and you have to do a new one after, and that 10 years is too much.

Who don't like to do things and like keep doing the same over and over again? That's right the company that he works(ed) for.
 
  • Like
Reactions: spursindonesia